{"domain":"","state":"","count":143,"shown":143,"sentences":125,"held":105,"merged":10,"total":258,"rows":[{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"TCL (Tracked Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"In light of this, we assess that we cannot rule out the TCL.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"In light of this, we assess that we cannot rule out the TCL.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-e5a149a655","anchor":"threshold-e5a149a655"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"We cannot rule out that Gemini 4 Argon reaches the biological CCL.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"We cannot rule out that Gemini 4 Argon reaches the biological CCL.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-7a5b67e27c","anchor":"threshold-7a5b67e27c"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"TCL (Tracked Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"We can rule out the TCL for the chemical, radiological, and nuclear domains with reasonable confidence.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"We can rule out the TCL for the chemical, radiological, and nuclear domains with reasonable confidence.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-0e21e502cc","anchor":"threshold-0e21e502cc"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"cyber","domain_label":"Cyber","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"We cannot rule out that Gemini 4 Argon reaches the cybersecurity CCL.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"We cannot rule out that Gemini 4 Argon reaches the cybersecurity CCL.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-a261637daf","anchor":"threshold-a261637daf"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess its overall capabilities to be below the CCL threshold.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"we assess its overall capabilities to be below the CCL threshold.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-6819cd7a4f","anchor":"threshold-6819cd7a4f"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Stealth and situational awareness TCL (Tracked Capability Level)","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"While Gemini 4 Argon exhibits capabilities that meet our thresholds for stealth and situational awareness, we assess that the risk of severe harm or unconstrained loss of control remains low.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"While Gemini 4 Argon exhibits capabilities that meet our thresholds for stealth and situational awareness, we assess that the risk of severe harm or unconstrained loss of control remains low.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-18a3dca6dd","anchor":"threshold-18a3dca6dd"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"misuse_and_jailbreak","domain_label":"Misuse & jailbreak","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"Gemini 4 Argon demonstrates some ability to influence user beliefs and behaviors during one-on-one direct conversations in human behavioral studies, and may provide utility in the planning stage of a malicious influence operation but we assess its overall capabilities to be below the CCL threshold.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"Gemini 4 Argon demonstrates some ability to influence user beliefs and behaviors during one-on-one direct conversations in human behavioral studies, and may provide utility in the planning stage of a malicious influence operation but we assess its overall capabilities to be below the CCL threshold.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-8775192ed9","anchor":"threshold-8775192ed9"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"Gemini 4 Argon's performance remains below our CCL threshold.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"Gemini 4 Argon's performance remains below our CCL threshold.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-0bc94628a5","anchor":"threshold-0bc94628a5"},{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Autonomy-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","publisher":"Anthropic","date":"7 October 2026","date_iso":"2026-10-07","url":"https://www.anthropic.com/claude-haiku-5-5-system-card","text":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0236#threshold-1f30090b25","anchor":"threshold-1f30090b25"},{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Autonomy-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","publisher":"Anthropic","date":"7 October 2026","date_iso":"2026-10-07","url":"https://www.anthropic.com/claude-haiku-5-5-system-card","text":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0236#threshold-800289c404","anchor":"threshold-800289c404"},{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","publisher":"Anthropic","date":"7 October 2026","date_iso":"2026-10-07","url":"https://www.anthropic.com/claude-haiku-5-5-system-card","text":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0236#threshold-588d3fff02","anchor":"threshold-588d3fff02"},{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","publisher":"Anthropic","date":"7 October 2026","date_iso":"2026-10-07","url":"https://www.anthropic.com/claude-haiku-5-5-system-card","text":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0236#threshold-df67b035be","anchor":"threshold-df67b035be"},{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Autonomy-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.","damaged":false,"lowered":true},"also":[{"quote":"as such, we re-assess the risks from this threat model in comparison to the analysis in our most recent risk report for all of our models capable enough to meet our autonomy-1 threshold of performing autonomous, goal-directed operation with access to sensitive resources (which includes claude sonnet 5.5), although our level of concern is generally lower for non-frontier models.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"as such, we re-assess the risks from this threat model in comparison to the analysis in our most recent risk report for all of our models capable enough to meet our autonomy-1 threshold of performing autonomous, goal-directed operation with access to sensitive resources (which includes claude sonnet 5.5), although our level of concern is generally lower for non-frontier models.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0226#threshold-bb00b7c97a","anchor":"threshold-bb00b7c97a"},{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Autonomy-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0226#threshold-a8b13d9a43","anchor":"threshold-a8b13d9a43"},{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0226#threshold-fe9ad9e5f5","anchor":"threshold-fe9ad9e5f5"},{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0226#threshold-5c0ef007b8","anchor":"threshold-5c0ef007b8"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model: (1) we do not observe a sustained ai-attributable 2× acceleration in the pace of our overall ai progress, and (2) the model is not close to substituting for our research scientists and research engineers, especially relatively senior ones.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model: (1) we do not observe a sustained ai-attributable 2× acceleration in the pace of our overall ai progress, and (2) the model is not close to substituting for our research scientists and research engineers, especially relatively senior ones.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-d92d3b6fe9","anchor":"threshold-d92d3b6fe9"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Dramatic acceleration of automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess that claude opus 5.5 does not cross the capability threshold for dramatic acceleration of automated ai r&d in our rsp.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"we assess that claude opus 5.5 does not cross the capability threshold for dramatic acceleration of automated ai r&d in our rsp.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-632abbddea","anchor":"threshold-632abbddea"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Next capability threshold in our RSP and FCF","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"in automated ai research and development, we assess that claude opus 5.5 does not cross the next capability threshold in our rsp and fcf.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"in automated ai research and development, we assess that claude opus 5.5 does not cross the next capability threshold in our rsp and fcf.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-d1bb9898f3","anchor":"threshold-d1bb9898f3"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"on chemical and biological risks, we treat opus 5.5 as having cb-1 capabilities (relating to the synthesis of non-novel weapons) but not cb-2 capabilities (relating to the synthesis of novel weapons).","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"on chemical and biological risks, we treat opus 5.5 as having cb-1 capabilities (relating to the synthesis of non-novel weapons) but not cb-2 capabilities (relating to the synthesis of novel weapons).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-7186b0e763","anchor":"threshold-7186b0e763"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"on chemical and biological risks, we treat opus 5.5 as having cb-1 capabilities (relating to the synthesis of non-novel weapons) but not cb-2 capabilities (relating to the synthesis of novel weapons).","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"on chemical and biological risks, we treat opus 5.5 as having cb-1 capabilities (relating to the synthesis of non-novel weapons) but not cb-2 capabilities (relating to the synthesis of novel weapons).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-0fe3b7d2de","anchor":"threshold-0fe3b7d2de"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"cyber","domain_label":"Cyber","level":"Lower category of capabilities set out in our frontier compliance framework","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"despite claude opus 5.5’s cyber capabilities, the model still falls within the lower category of capabilities set out in our frontier compliance framework (fcf), and we do not see indications that the model is capable of novel offensive capability.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"despite claude opus 5.5’s cyber capabilities, the model still falls within the lower category of capabilities set out in our frontier compliance framework (fcf), and we do not see indications that the model is capable of novel offensive capability.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-112c945031","anchor":"threshold-112c945031"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"cyber","domain_label":"Cyber","level":"Lower category of the Frontier Compliance Framework (FCF)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"despite claude opus 5.5’s cyber capabilities, the model still falls within the lower category of capabilities set out in our frontier compliance framework (fcf), and we do not see indications that the model is capable of novel offensive capability.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www.anthropic.com/claude-opus-5-5-system-card","text":"despite claude opus 5.5’s cyber capabilities, the model still falls within the lower category of capabilities set out in our frontier compliance framework (fcf), and we do not see indications that the model is capable of novel offensive capability.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-1382e7d589","anchor":"threshold-1382e7d589"},{"version":{"id":"MV-2026-0005","name":"Grok 4.7","provider":"xAI","provider_slug":"xai","released":"21 September 2026","released_iso":"2026-09-21","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0005"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"FAIF safety (Frontier AI Framework)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"grok 4.7 scores below the faif safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","publisher":"xAI","date":"21 September 2026","date_iso":"2026-09-21","url":"https://media.x.ai/v1/website/card4p7-3a96f40b.pdf","text":"grok 4.7 scores below the faif safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0005#threshold-904e31f7ee","anchor":"threshold-904e31f7ee"},{"version":{"id":"MV-2026-0014","name":"GPT-Image-2.5-Flare","provider":"OpenAI","provider_slug":"openai","released":"8 September 2026","released_iso":"2026-09-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0014"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the results indicate that neither gpt-image-2.5-sunburst nor gpt-image- 2.5-flare cross the bio high threshold or the cyber high threshold.","publisher":"OpenAI","date":"8 September 2026","date_iso":"2026-09-08","url":"https://deploymentsafety.openai.com/chatgpt-images-2-5/chatgpt-images-2-5.pdf","text":"the results indicate that neither gpt-image-2.5-sunburst nor gpt-image- 2.5-flare cross the bio high threshold or the cyber high threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0014#threshold-1e93ade02f","anchor":"threshold-1e93ade02f"},{"version":{"id":"MV-2026-0014","name":"GPT-Image-2.5-Flare","provider":"OpenAI","provider_slug":"openai","released":"8 September 2026","released_iso":"2026-09-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0014"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the results indicate that neither gpt-image-2.5-sunburst nor gpt-image- 2.5-flare cross the bio high threshold or the cyber high threshold.","publisher":"OpenAI","date":"8 September 2026","date_iso":"2026-09-08","url":"https://deploymentsafety.openai.com/chatgpt-images-2-5/chatgpt-images-2-5.pdf","text":"the results indicate that neither gpt-image-2.5-sunburst nor gpt-image- 2.5-flare cross the bio high threshold or the cyber high threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0014#threshold-9074d9d735","anchor":"threshold-9074d9d735"},{"version":{"id":"MV-2026-0020","name":"GPT-6 Astra","provider":"OpenAI","provider_slug":"openai","released":"3 September 2026","released_iso":"2026-09-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0020"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we are treating astra as high capability in the biological and chemical domain.","publisher":"OpenAI","date":"3 September 2026","date_iso":"2026-09-03","url":"https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf","text":"we are treating astra as high capability in the biological and chemical domain.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0020#threshold-976d12867e","anchor":"threshold-976d12867e"},{"version":{"id":"MV-2026-0020","name":"GPT-6 Astra","provider":"OpenAI","provider_slug":"openai","released":"3 September 2026","released_iso":"2026-09-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0020"},"domain":"cyber","domain_label":"Cyber","level":"Critical","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"based on our evaluations, we believe that astra meets the critical cybersecurity threshold.","publisher":"OpenAI","date":"3 September 2026","date_iso":"2026-09-03","url":"https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf","text":"based on our evaluations, we believe that astra meets the critical cybersecurity threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0020#threshold-fc9263e086","anchor":"threshold-fc9263e086"},{"version":{"id":"MV-2026-0025","name":"Claude Mythos 5.1","provider":"Anthropic","provider_slug":"anthropic","released":"1 September 2026","released_iso":"2026-09-01","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0025"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Capability threshold for dramatic acceleration of automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess that claude mythos 5.1 does not cross the capability threshold for dramatic acceleration of automated ai r&d in our rsp and fcf, although we think it provides a meaningful acceleration to ai r&d efforts in many circumstances, and is somewhat more capable in this domain than the models described in our previous system cards.","publisher":"Anthropic","date":"1 September 2026","date_iso":"2026-09-01","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card","text":"we assess that claude mythos 5.1 does not cross the capability threshold for dramatic acceleration of automated ai r&d in our rsp and fcf, although we think it provides a meaningful acceleration to ai r&d efforts in many circumstances, and is somewhat more capable in this domain than the models described in our previous system cards.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0025#threshold-2dad838ed9","anchor":"threshold-2dad838ed9"},{"version":{"id":"MV-2026-0025","name":"Claude Mythos 5.1","provider":"Anthropic","provider_slug":"anthropic","released":"1 September 2026","released_iso":"2026-09-01","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0025"},"domain":"cyber","domain_label":"Cyber","level":"TIER 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"although mythos 5.1 is in tier 1, it is getting closer to tier 2, completing more and more autonomous tasks.","publisher":"Anthropic","date":"1 September 2026","date_iso":"2026-09-01","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card","text":"although mythos 5.1 is in tier 1, it is getting closer to tier 2, completing more and more autonomous tasks.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0025#threshold-ee41cdfe3a","anchor":"threshold-ee41cdfe3a"},{"version":{"id":"MV-2026-0033","name":"Gemini 3.5 Audio","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 August 2026","url":"/safety/models/MV-2026-0033"},"domain":"other","domain_label":"Other","level":"Tracked or critical capability level (t/ccls)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"they did not reach any tracked or critical capability levels (t/ccls) outlined in our frontier safety framework .","publisher":"Google DeepMind","date":"26 August 2026","date_iso":"2026-08-26","url":"https://deepmind.google/models/model-cards/gemini-3-5-audio/","text":"they did not reach any tracked or critical capability levels (t/ccls) outlined in our frontier safety framework .","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0033#threshold-fadaf5631d","anchor":"threshold-fadaf5631d"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the model does not reach the ccl alert threshold.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"the model does not reach the ccl alert threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-e3c65e47b1","anchor":"threshold-e3c65e47b1"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we can rule out the ccl for the cbrn domain with reasonable confidence based on the results from our testing.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"we can rule out the ccl for the cbrn domain with reasonable confidence based on the results from our testing.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-aac36c2a6d","anchor":"threshold-aac36c2a6d"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"TCL (Tracked Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we can rule out the tcl for the cbrn domain with reasonable confidence based on the results from our testing.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"we can rule out the tcl for the cbrn domain with reasonable confidence based on the results from our testing.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-4d71a24b8a","anchor":"threshold-4d71a24b8a"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"cyber","domain_label":"Cyber","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"gemini 3.7 flash reaches the alert threshold for this ccl, but not the ccl.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"gemini 3.7 flash reaches the alert threshold for this ccl, but not the ccl.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-36b5e8b6f2","anchor":"threshold-36b5e8b6f2"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"TCL (Tracked Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the model does not reach the tcl.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"the model does not reach the tcl.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-4f1ae689a7","anchor":"threshold-4f1ae689a7"},{"version":{"id":"MV-2026-0048","name":"Grok 4.6","provider":"xAI","provider_slug":"xai","released":"12 August 2026","released_iso":"2026-08-12","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0048"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"FAIF safety (Frontier AI Framework)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"grok 4.6 scores below the faif safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","publisher":"xAI","date":"12 August 2026","date_iso":"2026-08-12","url":"https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf","text":"grok 4.6 scores below the faif safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0048#threshold-ebbd582fcd","anchor":"threshold-ebbd582fcd"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-2534ddcfa9","anchor":"threshold-2534ddcfa9"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower risk (inferred)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"- loss of control: moderate or lower risk (inferred).","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"- loss of control: moderate or lower risk (inferred).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-6ac2e0eeed","anchor":"threshold-6ac2e0eeed"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-20c7926cb4","anchor":"threshold-20c7926cb4"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-a4f8c43c06","anchor":"threshold-a4f8c43c06"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower risk (inferred)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"- cyber: moderate or lower risk (inferred);","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"- cyber: moderate or lower risk (inferred);","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-2b51266ba1","anchor":"threshold-2b51266ba1"},{"version":{"id":"MV-2026-0071","name":"Claude Opus 5","provider":"Anthropic","provider_slug":"anthropic","released":"24 July 2026","released_iso":"2026-07-24","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0071"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"on chemical and biological risks, we treat the model as having cb-1 capabilities (relating to the synthesis of non-novel weapons), but not cb-2 capabilities (relating to the synthesis of novel weapons).","publisher":"Anthropic","date":"24 July 2026","date_iso":"2026-07-24","url":"https://www-cdn.anthropic.com/b514064af1408018e64b1ad24e7d5e75850b4ffd/Claude%20Opus%205%20System%20Card.pdf","text":"on chemical and biological risks, we treat the model as having cb-1 capabilities (relating to the synthesis of non-novel weapons), but not cb-2 capabilities (relating to the synthesis of novel weapons).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0071#threshold-e902771bf0","anchor":"threshold-e902771bf0"},{"version":{"id":"MV-2026-0071","name":"Claude Opus 5","provider":"Anthropic","provider_slug":"anthropic","released":"24 July 2026","released_iso":"2026-07-24","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0071"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"on chemical and biological risks, we treat the model as having cb-1 capabilities (relating to the synthesis of non-novel weapons), but not cb-2 capabilities (relating to the synthesis of novel weapons).","publisher":"Anthropic","date":"24 July 2026","date_iso":"2026-07-24","url":"https://www-cdn.anthropic.com/b514064af1408018e64b1ad24e7d5e75850b4ffd/Claude%20Opus%205%20System%20Card.pdf","text":"on chemical and biological risks, we treat the model as having cb-1 capabilities (relating to the synthesis of non-novel weapons), but not cb-2 capabilities (relating to the synthesis of novel weapons).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0071#threshold-0ea0c6ea70","anchor":"threshold-0ea0c6ea70"},{"version":{"id":"MV-2026-0075","name":"Gemini 3.5 Flash-Lite","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"21 July 2026","url":"/safety/models/MV-2026-0075"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that, while gemini 3.5 flash-lite excels at agents and coding, it does not have meaningful new capabilities or material increases in performance with respect to the domains outlined in our frontier safety framework compared to gemini 3.1 pro; therefore, based on gemini 3.1 pro results, we are confident that gemini 3.5 flash-lite is also unlikely to reach any ccls.","publisher":"Google DeepMind","date":"21 July 2026","date_iso":"2026-07-21","url":"https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/","text":"our assessments have shown that, while gemini 3.5 flash-lite excels at agents and coding, it does not have meaningful new capabilities or material increases in performance with respect to the domains outlined in our frontier safety framework compared to gemini 3.1 pro; therefore, based on gemini 3.1 pro results, we are confident that gemini 3.5 flash-lite is also unlikely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0075#threshold-384b2b75e1","anchor":"threshold-384b2b75e1"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Capability checkpoint","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"muse spark 1.1 demonstrates improved coding capabilities, which are now meeting one of the indicative thresholds introduced in muse spark safety & preparedness report (specifically, resolving over half of unique tasks on swe-bench verified hard).","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"muse spark 1.1 demonstrates improved coding capabilities, which are now meeting one of the indicative thresholds introduced in muse spark safety & preparedness report (specifically, resolving over half of unique tasks on swe-bench verified hard).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-ae18f623b6","anchor":"threshold-ae18f623b6"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"in the loss of control domain, the model’s capabilities remain within the “moderate or lower” threshold under meta’s advanced ai scaling framework.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"in the loss of control domain, the model’s capabilities remain within the “moderate or lower” threshold under meta’s advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-1b3a6516f6","anchor":"threshold-1b3a6516f6"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments indicate that, when evaluated without mitigations applied, we cannot rule out muse spark 1.1’s capabilities meeting this threshold in both the chemical & biological and cybersecurity domains.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"our assessments indicate that, when evaluated without mitigations applied, we cannot rule out muse spark 1.1’s capabilities meeting this threshold in both the chemical & biological and cybersecurity domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-505fbbce0d","anchor":"threshold-505fbbce0d"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments indicate that, when evaluated without mitigations applied, we cannot rule out muse spark 1.1’s capabilities meeting this threshold in both the chemical & biological and cybersecurity domains.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"our assessments indicate that, when evaluated without mitigations applied, we cannot rule out muse spark 1.1’s capabilities meeting this threshold in both the chemical & biological and cybersecurity domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-894f659c86","anchor":"threshold-894f659c86"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"therefore, we conclude that we have addressed these risks from unmitigated muse spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under advanced ai scaling framework.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"therefore, we conclude that we have addressed these risks from unmitigated muse spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-6b5cfc2380","anchor":"threshold-6b5cfc2380"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"misuse_and_jailbreak","domain_label":"Misuse & jailbreak","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"therefore, we conclude that we have addressed these risks from unmitigated muse spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under our advanced ai scaling framework.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"therefore, we conclude that we have addressed these risks from unmitigated muse spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under our advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-ddc93b87b7","anchor":"threshold-ddc93b87b7"},{"version":{"id":"MV-2026-0099","name":"Claude Sonnet 5","provider":"Anthropic","provider_slug":"anthropic","released":"30 June 2026","released_iso":"2026-06-30","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0099"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"sonnet 5 does not cross the cb-2 threshold.","publisher":"Anthropic","date":"30 June 2026","date_iso":"2026-06-30","url":"https://www-cdn.anthropic.com/9e6a1044980d8c4ed85669faf9c2a8342e2e9f1e/Claude%20Sonnet%205%20System%20Card.pdf","text":"sonnet 5 does not cross the cb-2 threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0099#threshold-d8f02bfc4c","anchor":"threshold-d8f02bfc4c"},{"version":{"id":"MV-2026-0099","name":"Claude Sonnet 5","provider":"Anthropic","provider_slug":"anthropic","released":"30 June 2026","released_iso":"2026-06-30","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0099"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Notable capability","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"we consider performance on the virology tasks notable if: ● the model scores above 0.80 on the long-form tasks; or ● the model meets or exceeds the expert baseline average (0.221) on the multimodal task.","publisher":"Anthropic","date":"30 June 2026","date_iso":"2026-06-30","url":"https://www-cdn.anthropic.com/9e6a1044980d8c4ed85669faf9c2a8342e2e9f1e/Claude%20Sonnet%205%20System%20Card.pdf","text":"we consider performance on the virology tasks notable if: ● the model scores above 0.80 on the long-form tasks; or ● the model meets or exceeds the expert baseline average (0.221) on the multimodal task.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0099#threshold-c24df75a69","anchor":"threshold-c24df75a69"},{"version":{"id":"MV-2026-0100","name":"Claude Mythos 5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"9 June 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0100"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model, mythos preview.","publisher":"Anthropic","date":"9 June 2026","date_iso":"2026-06-09","url":"https://www.anthropic.com/claude-fable-5-mythos-5-system-card","text":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model, mythos preview.","damaged":false,"lowered":true},"also":[{"quote":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model, mythos preview.","publisher":"","date":"9 June 2026","date_iso":"2026-06-09","url":"https://anthropic.com/claude-fable-5-mythos-5-system-card","text":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model, mythos preview.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0100#threshold-dee73e4d13","anchor":"threshold-dee73e4d13"},{"version":{"id":"MV-2026-0109","name":"Gemini 3.5 Transcribe","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 June 2026","url":"/safety/models/MV-2026-0109"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that none of gemini 3.5 translate, gemini 3.5 transcribe, or gemini 3.5 transcribe live have meaningful new t/ccl-relevant capabilities or material increases in performance with respect to frontier safety compared to gemini 3.1 pro and gemini 3.7 flash, therefore we are confident that none of them are likely to reach any ccls.","publisher":"Google DeepMind","date":"9 June 2026","date_iso":"2026-06-09","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-5-Audio-Model-Card.pdf","text":"our assessments have shown that none of gemini 3.5 translate, gemini 3.5 transcribe, or gemini 3.5 transcribe live have meaningful new t/ccl-relevant capabilities or material increases in performance with respect to frontier safety compared to gemini 3.1 pro and gemini 3.7 flash, therefore we are confident that none of them are likely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0109#threshold-49960a3c41","anchor":"threshold-49960a3c41"},{"version":{"id":"MV-2026-0110","name":"Gemini 3.5 Transcribe Live","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 June 2026","url":"/safety/models/MV-2026-0110"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that none of gemini 3.5 translate, gemini 3.5 transcribe, or gemini 3.5 transcribe live have meaningful new t/ccl-relevant capabilities or material increases in performance with respect to frontier safety compared to gemini 3.1 pro and gemini 3.7 flash, therefore we are confident that none of them are likely to reach any ccls.","publisher":"Google DeepMind","date":"9 June 2026","date_iso":"2026-06-09","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-5-Audio-Model-Card.pdf","text":"our assessments have shown that none of gemini 3.5 translate, gemini 3.5 transcribe, or gemini 3.5 transcribe live have meaningful new t/ccl-relevant capabilities or material increases in performance with respect to frontier safety compared to gemini 3.1 pro and gemini 3.7 flash, therefore we are confident that none of them are likely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0110#threshold-756533ff5d","anchor":"threshold-756533ff5d"},{"version":{"id":"MV-2026-0114","name":"GPT-Rosalind-5.5","provider":"OpenAI","provider_slug":"openai","released":"3 June 2026","released_iso":"2026-06-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0114"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"because gpt- rosalind-5.5 was trained on additional biology data unrelated to ai self-improvement, we rely on gpt-5.5’s existing preparedness assessment for ai self-improvement, where gpt-5.5 did not meet the threshold for high capability.","publisher":"OpenAI","date":"3 June 2026","date_iso":"2026-06-03","url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","text":"because gpt- rosalind-5.5 was trained on additional biology data unrelated to ai self-improvement, we rely on gpt-5.5’s existing preparedness assessment for ai self-improvement, where gpt-5.5 did not meet the threshold for high capability.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0114#threshold-5bd390c4a0","anchor":"threshold-5bd390c4a0"},{"version":{"id":"MV-2026-0114","name":"GPT-Rosalind-5.5","provider":"OpenAI","provider_slug":"openai","released":"3 June 2026","released_iso":"2026-06-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0114"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"under our preparedness framework, the preparedness evaluations in the biological and chemical domain met our threshold for high capability while falling below the threshold for critical.","publisher":"OpenAI","date":"3 June 2026","date_iso":"2026-06-03","url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","text":"under our preparedness framework, the preparedness evaluations in the biological and chemical domain met our threshold for high capability while falling below the threshold for critical.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0114#threshold-1445fc72e4","anchor":"threshold-1445fc72e4"},{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"loss of control risks.we assess loss of control risk as “moderate or lower.”","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"loss of control risks.we assess loss of control risk as “moderate or lower.”","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0115#threshold-34d87b7657","anchor":"threshold-34d87b7657"},{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we retain the same “high risk” determination for the unmitigated model.","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"we retain the same “high risk” determination for the unmitigated model.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0115#threshold-0e70889ea3","anchor":"threshold-0e70889ea3"},{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"after validating that the mitigations deployed for muse spark, including refusal mechanisms, continue to perform robustly under the contemplating architecture, we confirm residual risk at “moderate or lower.”","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"after validating that the mitigations deployed for muse spark, including refusal mechanisms, continue to perform robustly under the contemplating architecture, we confirm residual risk at “moderate or lower.”","damaged":false,"lowered":true},"also":[{"quote":"after applying the same mitigations as for muse spark, the residual risk is reduced to “moderate or lower.”","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"after applying the same mitigations as for muse spark, the residual risk is reduced to “moderate or lower.”","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0115#threshold-c251b7dd0f","anchor":"threshold-c251b7dd0f"},{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in meta’s advanced ai scaling framework.","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"we assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in meta’s advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0115#threshold-1696252301","anchor":"threshold-1696252301"},{"version":{"id":"MV-2026-0098","name":"Gemini 3.1 Flash-Lite Image","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"June 2026","released_iso":"2026-06","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0098"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that gemini 3.1 flash-lite image is less capable than gemini 3.1 pro, therefore based on gemini 3.1 pro, we are confident that that gemini 3.1 flash-lite image is also unlikely to reach any ccls.","publisher":"Google DeepMind","date":"June 2026","date_iso":"2026-06","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Flash-Lite-Image-Model-Card.pdf","text":"our assessments have shown that gemini 3.1 flash-lite image is less capable than gemini 3.1 pro, therefore based on gemini 3.1 pro, we are confident that that gemini 3.1 flash-lite image is also unlikely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0098#threshold-4fd12b5ca1","anchor":"threshold-4fd12b5ca1"},{"version":{"id":"MV-2026-0125","name":"Claude Opus 4.8","provider":"Anthropic","provider_slug":"anthropic","released":"28 May 2026","released_iso":"2026-05-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0125"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess that claude opus 4.8 does not cross the automated ai-r&d capability threshold.","publisher":"Anthropic","date":"28 May 2026","date_iso":"2026-05-28","url":"https://www.anthropic.com/claude-opus-4-8-system-card","text":"we assess that claude opus 4.8 does not cross the automated ai-r&d capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0125#threshold-c127a687aa","anchor":"threshold-c127a687aa"},{"version":{"id":"MV-2026-0125","name":"Claude Opus 4.8","provider":"Anthropic","provider_slug":"anthropic","released":"28 May 2026","released_iso":"2026-05-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0125"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"non-novel chemical and biological weapons (cb-1): we believe it is hard to be confident whether a model passes this threshold, but (as with previous recent models) our results are consistent with opus 4.8 providing significant uplift to individuals or groups with basic technical backgrounds for this threat model.","publisher":"Anthropic","date":"28 May 2026","date_iso":"2026-05-28","url":"https://www.anthropic.com/claude-opus-4-8-system-card","text":"non-novel chemical and biological weapons (cb-1): we believe it is hard to be confident whether a model passes this threshold, but (as with previous recent models) our results are consistent with opus 4.8 providing significant uplift to individuals or groups with basic technical backgrounds for this threat model.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0125#threshold-3d542dcc6a","anchor":"threshold-3d542dcc6a"},{"version":{"id":"MV-2026-0125","name":"Claude Opus 4.8","provider":"Anthropic","provider_slug":"anthropic","released":"28 May 2026","released_iso":"2026-05-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0125"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"because opus 4.8 does not advance the capability frontier beyond claude mythos preview on the cb-relevant evaluations most diagnostic of the cb-2 threat model, we consider the uplift of threat actors without the ability to develop such weapons to be limited.","publisher":"Anthropic","date":"28 May 2026","date_iso":"2026-05-28","url":"https://www.anthropic.com/claude-opus-4-8-system-card","text":"because opus 4.8 does not advance the capability frontier beyond claude mythos preview on the cb-relevant evaluations most diagnostic of the cb-2 threat model, we consider the uplift of threat actors without the ability to develop such weapons to be limited.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0125#threshold-e5a6c69ce6","anchor":"threshold-e5a6c69ce6"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess loss of control risk as “moderate or lower”.","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"we assess loss of control risk as “moderate or lower”.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-f1da2a69f4","anchor":"threshold-f1da2a69f4"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"those evaluations identified elevated risks prior to mitigations, with chemical and biological capabilities assessed as likely reaching the “high risk” category under the advanced ai scaling framework before safeguards were applied.","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"those evaluations identified elevated risks prior to mitigations, with chemical and biological capabilities assessed as likely reaching the “high risk” category under the advanced ai scaling framework before safeguards were applied.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-4c6c2a4863","anchor":"threshold-4c6c2a4863"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"however, after applying mitigations, the residual risk of the model is that of “moderate or lower risk.”","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"however, after applying mitigations, the residual risk of the model is that of “moderate or lower risk.”","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-d040130bc3","anchor":"threshold-d040130bc3"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"cyber","domain_label":"Cyber","level":"Below the level required to meaningfully enable the cyber threat scenarios defin","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"muse spark’s offensive cyber capabilities remain below the level required to meaningfully enable the cyber threat scenarios defined in the advanced ai scaling framework.","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"muse spark’s offensive cyber capabilities remain below the level required to meaningfully enable the cyber threat scenarios defined in the advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-3bb6399c37","anchor":"threshold-3bb6399c37"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in meta’s advanced ai scaling framework.","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"we assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in meta’s advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-19a0b4617e","anchor":"threshold-19a0b4617e"},{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Below high capability","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"gpt-5.5 instant is considered below high capability.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"gpt-5.5 instant is considered below high capability.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0135#threshold-5b6a76027d","anchor":"threshold-5b6a76027d"},{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.5 instant does not have a plausible chance of reaching a high threshold.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.5 instant does not have a plausible chance of reaching a high threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0135#threshold-78bb6c1440","anchor":"threshold-78bb6c1440"},{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"gpt-5.5 instant is our first instant model to be treated as high capability in the biological and chemical domain, as well as in the cybersecurity domain.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"gpt-5.5 instant is our first instant model to be treated as high capability in the biological and chemical domain, as well as in the cybersecurity domain.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0135#threshold-b623243f6e","anchor":"threshold-b623243f6e"},{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"this is the first instant model that we are treating as high capability in our cybersecurity and biological & chemical preparedness categories, and implementing appropriate safeguards.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"this is the first instant model that we are treating as high capability in our cybersecurity and biological & chemical preparedness categories, and implementing appropriate safeguards.","damaged":false,"lowered":true},"also":[{"quote":"because of this, we similarly treat gpt-5.5 instant as high under the preparedness framework.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"because of this, we similarly treat gpt-5.5 instant as high under the preparedness framework.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0135#threshold-06d06fe664","anchor":"threshold-06d06fe664"},{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.5 does not have a plausible chance of reaching a high threshold.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.5 does not have a plausible chance of reaching a high threshold.","damaged":false,"lowered":true},"also":[{"quote":"gpt-5.5 did not meet our thresholds for high capability in ai self-improvement.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"gpt-5.5 did not meet our thresholds for high capability in ai self-improvement.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0154#threshold-cfe2b3524c","anchor":"threshold-cfe2b3524c"},{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we are continuing to treat gpt-5.5 as high capability in the biological and chemical domain.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"we are continuing to treat gpt-5.5 as high capability in the biological and chemical domain.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0154#threshold-5ea957e9c3","anchor":"threshold-5ea957e9c3"},{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"domain":"cyber","domain_label":"Cyber","level":"Critical","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we therefore treat vulnlmp as important evidence that gpt-5.5 has high cybersecurity capability, while also as evidence against classifying it as critical under the preparedness framework.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"we therefore treat vulnlmp as important evidence that gpt-5.5 has high cybersecurity capability, while also as evidence against classifying it as critical under the preparedness framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0154#threshold-3d972a74f4","anchor":"threshold-3d972a74f4"},{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"domain":"cyber","domain_label":"Cyber","level":"High capability, below critical","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we are treating gpt-5.5 as high capability in the cybersecurity domain, but below critical.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"we are treating gpt-5.5 as high capability in the cybersecurity domain, but below critical.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0154#threshold-ad63b86a7d","anchor":"threshold-ad63b86a7d"},{"version":{"id":"MV-2026-0167","name":"Claude Opus 4.7","provider":"Anthropic","provider_slug":"anthropic","released":"16 April 2026","released_iso":"2026-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0167"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"that opus 4.7 does not cross the threshold for automated ai r&d;","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://www.anthropic.com/claude-opus-4-7-system-card","text":"that opus 4.7 does not cross the threshold for automated ai r&d;","damaged":false,"lowered":true},"also":[{"quote":"we assess that claude opus 4.7 does not cross the automated ai-r&d capability threshold.","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://www.anthropic.com/claude-opus-4-7-system-card","text":"we assess that claude opus 4.7 does not cross the automated ai-r&d capability threshold.","damaged":false,"lowered":true},{"quote":"that opus 4.7 does not cross the threshold for automated ai r&d; and that misalignment risk remains very low (though higher than for pre-mythos preview models).","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://anthropic.com/claude-opus-4-7-system-card","text":"that opus 4.7 does not cross the threshold for automated ai r&d; and that misalignment risk remains very low (though higher than for pre-mythos preview models).","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0167#threshold-1d015e9fdb","anchor":"threshold-1d015e9fdb"},{"version":{"id":"MV-2026-0167","name":"Claude Opus 4.7","provider":"Anthropic","provider_slug":"anthropic","released":"16 April 2026","released_iso":"2026-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0167"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"we believe it is hard to be confident regarding whether a model passes this threshold.","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://www.anthropic.com/claude-opus-4-7-system-card","text":"we believe it is hard to be confident regarding whether a model passes this threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0167#threshold-09307b504a","anchor":"threshold-09307b504a"},{"version":{"id":"MV-2026-0167","name":"Claude Opus 4.7","provider":"Anthropic","provider_slug":"anthropic","released":"16 April 2026","released_iso":"2026-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0167"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our evaluations suggest that the model is not yet at the level of capability associated with the cb-2 threat model (above).","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://www.anthropic.com/claude-opus-4-7-system-card","text":"our evaluations suggest that the model is not yet at the level of capability associated with the cb-2 threat model (above).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0167#threshold-8fe8052e3a","anchor":"threshold-8fe8052e3a"},{"version":{"id":"MV-2026-0170","name":"Gemini 3.1 Flash Audio","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"15 April 2026","url":"/safety/models/MV-2026-0170"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that gemini 3.1 flash audio is less capable than gemini 3.1 pro.","publisher":"Google DeepMind","date":"15 April 2026","date_iso":"2026-04-15","url":"https://deepmind.google/models/model-cards/gemini-3-1-flash-audio/","text":"our assessments have shown that gemini 3.1 flash audio is less capable than gemini 3.1 pro.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0170#threshold-7829dbbc8f","anchor":"threshold-7829dbbc8f"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D rule-out","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the main reason we have determined that claude mythos preview does not cross the threshold in question is that we have been using it extensively in the course of our day-to-day work and exploring where it can automate such work, and it does not seem close to being able to substitute for research scientists and research engineers—especially relatively senior ones.","publisher":"","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www.anthropic.com/claude-mythos-preview-system-card","text":"the main reason we have determined that claude mythos preview does not cross the threshold in question is that we have been using it extensively in the course of our day-to-day work and exploring where it can automate such work, and it does not seem close to being able to substitute for research scientists and research engineers—especially relatively senior ones.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0181#threshold-78837e5e82","anchor":"threshold-78837e5e82"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess that claude mythos preview does not cross the automated ai-r&d capability threshold.","publisher":"Anthropic","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","text":"we assess that claude mythos preview does not cross the automated ai-r&d capability threshold.","damaged":false,"lowered":true},"also":[{"quote":"we assess that claude mythos preview does not cross the automated ai-r&d capability threshold.","publisher":"","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www.anthropic.com/claude-mythos-preview-system-card","text":"we assess that claude mythos preview does not cross the automated ai-r&d capability threshold.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0181#threshold-5c424f12f2","anchor":"threshold-5c424f12f2"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"RSP automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we have concluded that claude mythos preview does not cross the rsp automated ai r&d threshold of compressing two years of progress into one.","publisher":"Anthropic","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","text":"we have concluded that claude mythos preview does not cross the rsp automated ai r&d threshold of compressing two years of progress into one.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0181#threshold-fbcd7989d6","anchor":"threshold-fbcd7989d6"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"we believe it is hard to be confident regarding whether a model passes this threshold.","publisher":"Anthropic","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","text":"we believe it is hard to be confident regarding whether a model passes this threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0181#threshold-70183ff1fb","anchor":"threshold-70183ff1fb"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we believe that claude mythos preview does not pass this threshold due to its noted limitations in open-ended scientific reasoning, strategic judgment, and hypothesis triage.","publisher":"Anthropic","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","text":"we believe that claude mythos preview does not pass this threshold due to its noted limitations in open-ended scientific reasoning, strategic judgment, and hypothesis triage.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0181#threshold-a224925ca4","anchor":"threshold-a224925ca4"},{"version":{"id":"MV-2026-0200","name":"GPT-5.4 Thinking","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0200"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.4-thinking does not have a plausible chance of reaching a high threshold.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.4-thinking does not have a plausible chance of reaching a high threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0200#threshold-69cf502d31","anchor":"threshold-69cf502d31"},{"version":{"id":"MV-2026-0200","name":"GPT-5.4 Thinking","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0200"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"as we did for gpt-5.2-thinking before it, we are continuing to treat gpt-5.4-thinking as high capability in the biological and chemical domain.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"as we did for gpt-5.2-thinking before it, we are continuing to treat gpt-5.4-thinking as high capability in the biological and chemical domain.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0200#threshold-5fc91c31d0","anchor":"threshold-5fc91c31d0"},{"version":{"id":"MV-2026-0200","name":"GPT-5.4 Thinking","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0200"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"as we did for gpt-5.3-codex, we are treating gpt-5.4-thinking as high capability in the cybersecurity domain, and applied safeguards as described in the safeguard section below.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"as we did for gpt-5.3-codex, we are treating gpt-5.4-thinking as high capability in the cybersecurity domain, and applied safeguards as described in the safeguard section below.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0200#threshold-a12e61e384","anchor":"threshold-a12e61e384"},{"version":{"id":"MV-2026-0201","name":"GPT-5.4 mini","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0201"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"gpt-5.4 mini’s biology capabilities are closely comparable to those of gpt-5. we conclude that it, too, is not high under our preparedness framework.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"gpt-5.4 mini’s biology capabilities are closely comparable to those of gpt-5. we conclude that it, too, is not high under our preparedness framework.","damaged":false,"lowered":true},"also":[{"quote":"we therefore conclude that gpt-5.4 mini should not be classified as bio high under the current preparedness framework novice-uplift criterion.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"we therefore conclude that gpt-5.4 mini should not be classified as bio high under the current preparedness framework novice-uplift criterion.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0201#threshold-762f00523f","anchor":"threshold-762f00523f"},{"version":{"id":"MV-2026-0203","name":"Gemini 3.1 Flash-Lite","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 March 2026","url":"/safety/models/MV-2026-0203"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that gemini 3.1 flash-lite is less capable than gemini 3.1 pro, therefore based on gemini 3.1 pro, we are confident that gemini 3.1 flash-lite is also unlikely to reach any ccls.","publisher":"Google DeepMind","date":"3 March 2026","date_iso":"2026-03-03","url":"https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/","text":"our assessments have shown that gemini 3.1 flash-lite is less capable than gemini 3.1 pro, therefore based on gemini 3.1 pro, we are confident that gemini 3.1 flash-lite is also unlikely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0203#threshold-4bde6465d5","anchor":"threshold-4bde6465d5"},{"version":{"id":"MV-2026-0205","name":"Gemini 3.1 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"19 February 2026","url":"/safety/models/MV-2026-0205"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Uplift level 1","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we found that the model remains below alert thresholds for the cbrn, harmful manipulation, machine learning r&d, and misalignment ccls.","publisher":"Google DeepMind","date":"19 February 2026","date_iso":"2026-02-19","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Pro-Model-Card.pdf","text":"we found that the model remains below alert thresholds for the cbrn, harmful manipulation, machine learning r&d, and misalignment ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0205#threshold-3fd67b7dac","anchor":"threshold-3fd67b7dac"},{"version":{"id":"MV-2026-0205","name":"Gemini 3.1 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"19 February 2026","url":"/safety/models/MV-2026-0205"},"domain":"cyber","domain_label":"Cyber","level":"Uplift level 1","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"as previous models passed the alert threshold for cyber, we performed more additional testing in this domain on gemini 3.1 pro with and without deep think mode, and found that the model remains below the cyber ccl.","publisher":"Google DeepMind","date":"19 February 2026","date_iso":"2026-02-19","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Pro-Model-Card.pdf","text":"as previous models passed the alert threshold for cyber, we performed more additional testing in this domain on gemini 3.1 pro with and without deep think mode, and found that the model remains below the cyber ccl.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0205#threshold-9c774352dc","anchor":"threshold-9c774352dc"},{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude sonnet 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","publisher":"Anthropic","date":"17 February 2026","date_iso":"2026-02-17","url":"https://www.anthropic.com/claude-sonnet-4-6-system-card","text":"our determination is that claude sonnet 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0208#threshold-6b840577c6","anchor":"threshold-6b840577c6"},{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"claude sonnet 4.6 performed above thresholds for asl-3 rule-in on all three evaluations for asl-3, and it did not exceed the performance of previous models.","publisher":"Anthropic","date":"17 February 2026","date_iso":"2026-02-17","url":"https://www.anthropic.com/claude-sonnet-4-6-system-card","text":"claude sonnet 4.6 performed above thresholds for asl-3 rule-in on all three evaluations for asl-3, and it did not exceed the performance of previous models.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0208#threshold-e6b4af94f2","anchor":"threshold-e6b4af94f2"},{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"in particular, it did not cross the threshold on our asl-4 rule-out evaluation short-horizon computational biology tasks evaluation.","publisher":"Anthropic","date":"17 February 2026","date_iso":"2026-02-17","url":"https://www.anthropic.com/claude-sonnet-4-6-system-card","text":"in particular, it did not cross the threshold on our asl-4 rule-out evaluation short-horizon computational biology tasks evaluation.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0208#threshold-aa23cfe1d1","anchor":"threshold-aa23cfe1d1"},{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CBRN-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude sonnet 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","publisher":"Anthropic","date":"17 February 2026","date_iso":"2026-02-17","url":"https://www.anthropic.com/claude-sonnet-4-6-system-card","text":"our determination is that claude sonnet 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0208#threshold-6104bf9d7d","anchor":"threshold-6104bf9d7d"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude opus 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"our determination is that claude opus 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-0012ab64b5","anchor":"threshold-0012ab64b5"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-4","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"for ai r&d capabilities, we found that claude opus 4.6 has saturated most of our automated evaluations, meaning they no longer provide useful evidence for ruling out asl-4 level autonomy.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"for ai r&d capabilities, we found that claude opus 4.6 has saturated most of our automated evaluations, meaning they no longer provide useful evidence for ruling out asl-4 level autonomy.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-b06b179120","anchor":"threshold-b06b179120"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"the model crossed or met thresholds on all asl-3 evaluations except our synthesis screening evasion, consistent with incremental capability improvements driven primarily by better agentic workflows.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"the model crossed or met thresholds on all asl-3 evaluations except our synthesis screening evasion, consistent with incremental capability improvements driven primarily by better agentic workflows.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-e24763e7fa","anchor":"threshold-e24763e7fa"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"claude opus 4.6 is our strongest biology model to date and demonstrates meaningful capability improvements, but we do not believe it merits asl-4 safeguards.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"claude opus 4.6 is our strongest biology model to date and demonstrates meaningful capability improvements, but we do not believe it merits asl-4 safeguards.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-556fc6a3bb","anchor":"threshold-556fc6a3bb"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CBRN-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude opus 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"our determination is that claude opus 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-32187b620a","anchor":"threshold-32187b620a"},{"version":{"id":"MV-2026-0216","name":"GPT-5.3-Codex","provider":"OpenAI","provider_slug":"openai","released":"5 February 2026","released_iso":"2026-02-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0216"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"it does not reach high capability on ai self-improvement.","publisher":"OpenAI","date":"5 February 2026","date_iso":"2026-02-05","url":"https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf","text":"it does not reach high capability on ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0216#threshold-dfa617b4dc","anchor":"threshold-dfa617b4dc"},{"version":{"id":"MV-2026-0216","name":"GPT-5.3-Codex","provider":"OpenAI","provider_slug":"openai","released":"5 February 2026","released_iso":"2026-02-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0216"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we are also treating gpt-5.3-codex as high risk in the biological and chemical domain, and applying the same safeguards as for our other deployments of models that reach this designation.","publisher":"OpenAI","date":"5 February 2026","date_iso":"2026-02-05","url":"https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf","text":"we are also treating gpt-5.3-codex as high risk in the biological and chemical domain, and applying the same safeguards as for our other deployments of models that reach this designation.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0216#threshold-5bccffb03a","anchor":"threshold-5bccffb03a"},{"version":{"id":"MV-2026-0216","name":"GPT-5.3-Codex","provider":"OpenAI","provider_slug":"openai","released":"5 February 2026","released_iso":"2026-02-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0216"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"this is the first launch we are treating as high capability in the cybersecurity domain under our preparedness framework, activating the associated safeguards.","publisher":"OpenAI","date":"5 February 2026","date_iso":"2026-02-05","url":"https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf","text":"this is the first launch we are treating as high capability in the cybersecurity domain under our preparedness framework, activating the associated safeguards.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0216#threshold-3022fdc78e","anchor":"threshold-3022fdc78e"},{"version":{"id":"MV-2025-0005","name":"Claude Opus 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"November 2025","url":"/safety/models/MV-2025-0005"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude opus 4.5 does not cross either the ai r&d-4 or cbrn-4 capability threshold.","publisher":"Anthropic","date":"November 2025","date_iso":"2025-11","url":"https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf","text":"our determination is that claude opus 4.5 does not cross either the ai r&d-4 or cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0005#threshold-8bc97337aa","anchor":"threshold-8bc97337aa"},{"version":{"id":"MV-2025-0005","name":"Claude Opus 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"November 2025","url":"/safety/models/MV-2025-0005"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CBRN-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we determine that claude opus 4.5 does not cross the cbrn-4 threshold.","publisher":"Anthropic","date":"November 2025","date_iso":"2025-11","url":"https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf","text":"we determine that claude opus 4.5 does not cross the cbrn-4 threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0005#threshold-c188f9c9cf","anchor":"threshold-c188f9c9cf"},{"version":{"id":"MV-2025-0004","name":"Gemini 3 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"November 2025","released_iso":"2025-11","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0004"},"domain":"cyber","domain_label":"Cyber","level":"Uplift level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"cybersecurity on key skills benchmark, v1 hard challenges: 11/12 challenges solved; v2 challenges: 0/13 solved end-to-end.","publisher":"Google DeepMind","date":"November 2025","date_iso":"2025-11","url":"https://deepmind.google/models/model-cards/gemini-3-pro/","text":"cybersecurity on key skills benchmark, v1 hard challenges: 11/12 challenges solved; v2 challenges: 0/13 solved end-to-end.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0004#threshold-80aeac0c5c","anchor":"threshold-80aeac0c5c"},{"version":{"id":"MV-2025-0004","name":"Gemini 3 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"November 2025","released_iso":"2025-11","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0004"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Instrumental reasoning level 1 + 2","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"misalignment (exploratory) agent solves 3/11 situational awareness challenges and 1/4 stealth challenges.","publisher":"Google DeepMind","date":"November 2025","date_iso":"2025-11","url":"https://deepmind.google/models/model-cards/gemini-3-pro/","text":"misalignment (exploratory) agent solves 3/11 situational awareness challenges and 1/4 stealth challenges.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0004#threshold-03acb1b1a4","anchor":"threshold-03acb1b1a4"},{"version":{"id":"MV-2025-0004","name":"Gemini 3 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"November 2025","released_iso":"2025-11","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0004"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Level 1 (exploratory)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"harmful manipulation model manipulative efficacy improves on non-generative ai baseline, but shows no significant uplift versus prior models and does not reach alert thresholds.","publisher":"Google DeepMind","date":"November 2025","date_iso":"2025-11","url":"https://deepmind.google/models/model-cards/gemini-3-pro/","text":"harmful manipulation model manipulative efficacy improves on non-generative ai baseline, but shows no significant uplift versus prior models and does not reach alert thresholds.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0004#threshold-a5082992ed","anchor":"threshold-a5082992ed"},{"version":{"id":"MV-2025-0007","name":"Claude Haiku 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"October 2025","url":"/safety/models/MV-2025-0007"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-3","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our evaluation results showed that claude haiku 4.5 remained well below asl-3 thresholds across all domains of concern.","publisher":"Anthropic","date":"October 2025","date_iso":"2025-10","url":"https://www-cdn.anthropic.com/7aad69bf12627d42234e01ee7c36305dc2f6a970.pdf","text":"our evaluation results showed that claude haiku 4.5 remained well below asl-3 thresholds across all domains of concern.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0007#threshold-388ca4765f","anchor":"threshold-388ca4765f"},{"version":{"id":"MV-2025-0007","name":"Claude Haiku 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"October 2025","url":"/safety/models/MV-2025-0007"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our evaluation results showed that claude haiku 4.5 remained well below asl-3 thresholds across all domains of concern.","publisher":"Anthropic","date":"October 2025","date_iso":"2025-10","url":"https://www-cdn.anthropic.com/7aad69bf12627d42234e01ee7c36305dc2f6a970.pdf","text":"our evaluation results showed that claude haiku 4.5 remained well below asl-3 thresholds across all domains of concern.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0007#threshold-04f4ae624a","anchor":"threshold-04f4ae624a"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"all models tested remain below our pre-committed rule-out threshold of 0.6.","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"all models tested remain below our pre-committed rule-out threshold of 0.6.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-b48be0f714","anchor":"threshold-b48be0f714"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"exhibited improved autonomy capabilities in software engineering and ai research tasks, though still below the asl-4 rule-out threshold.","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"exhibited improved autonomy capabilities in software engineering and ai research tasks, though still below the asl-4 rule-out threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-66a676ff96","anchor":"threshold-66a676ff96"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"remained well below asl-4 thresholds across all domains of concern;","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"remained well below asl-4 thresholds across all domains of concern;","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-21b885ffab","anchor":"threshold-21b885ffab"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4 rule-out","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"however, given that this type of logic can have a general frog-boiling effect—where gradual changes mask the crossing of important thresholds—we do not use this evaluation as indicative of a strong asl-4 rule-out.","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"however, given that this type of logic can have a general frog-boiling effect—where gradual changes mask the crossing of important thresholds—we do not use this evaluation as indicative of a strong asl-4 rule-out.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-c68b5f8874","anchor":"threshold-c68b5f8874"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"cyber","domain_label":"Cyber","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"showed meaningful improvements in cyber capabilities, particularly in vulnerability discovery and code analysis;","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"showed meaningful improvements in cyber capabilities, particularly in vulnerability discovery and code analysis;","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-893d0f3495","anchor":"threshold-893d0f3495"},{"version":{"id":"MV-2025-0014","name":"GPT-5","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"13 August 2025","provider_doc_dated":"","url":"/safety/models/MV-2025-0014"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"similarly to chatgpt agent, we have decided to treat gpt-5-thinking as high capability in the biological and chemical domain under our preparedness framework, activating the associated safeguards.","publisher":"OpenAI","date":"13 August 2025","date_iso":"2025-08-13","url":"https://cdn.openai.com/gpt-5-system-card.pdf","text":"similarly to chatgpt agent, we have decided to treat gpt-5-thinking as high capability in the biological and chemical domain under our preparedness framework, activating the associated safeguards.","damaged":false,"lowered":true},"also":[{"quote":"we decided to treat this launch as high capability in the biological and chemical domain, activating the associated preparedness safeguards.","publisher":"OpenAI","date":"13 August 2025","date_iso":"2025-08-13","url":"https://cdn.openai.com/gpt-5-system-card.pdf","text":"we decided to treat this launch as high capability in the biological and chemical domain, activating the associated preparedness safeguards.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2025-0014#threshold-f2c6b00a38","anchor":"threshold-f2c6b00a38"},{"version":{"id":"MV-2025-0014","name":"GPT-5","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"13 August 2025","provider_doc_dated":"","url":"/safety/models/MV-2025-0014"},"domain":"cyber","domain_label":"Cyber","level":"High cyber risk","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the gpt-5 model series does not meet the threshold for high cyber risk.","publisher":"OpenAI","date":"13 August 2025","date_iso":"2025-08-13","url":"https://cdn.openai.com/gpt-5-system-card.pdf","text":"the gpt-5 model series does not meet the threshold for high cyber risk.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0014#threshold-59ada166b8","anchor":"threshold-59ada166b8"},{"version":{"id":"MV-2025-0017","name":"Claude Opus 4.1","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0017"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","publisher":"Anthropic","date":"August 2025","date_iso":"2025-08","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf","text":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0017#threshold-eb476b7fde","anchor":"threshold-eb476b7fde"},{"version":{"id":"MV-2025-0017","name":"Claude Opus 4.1","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0017"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","publisher":"Anthropic","date":"August 2025","date_iso":"2025-08","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf","text":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0017#threshold-0d0967fc34","anchor":"threshold-0d0967fc34"},{"version":{"id":"MV-2025-0017","name":"Claude Opus 4.1","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0017"},"domain":"cyber","domain_label":"Cyber","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","publisher":"Anthropic","date":"August 2025","date_iso":"2025-08","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf","text":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0017#threshold-62c30509fe","anchor":"threshold-62c30509fe"},{"version":{"id":"MV-2025-0016","name":"Gemini 2.5 Deep Think","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0016"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Uplift level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessment is that the model has enough technical knowledge in certain cbrn scenarios and stages to be considered at early warning.","publisher":"Google DeepMind","date":"August 2025","date_iso":"2025-08","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf","text":"our assessment is that the model has enough technical knowledge in certain cbrn scenarios and stages to be considered at early warning.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0016#threshold-ad43754209","anchor":"threshold-ad43754209"},{"version":{"id":"MV-2025-0016","name":"Gemini 2.5 Deep Think","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0016"},"domain":"cyber","domain_label":"Cyber","level":"Cyber Uplift Level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"ccl reached? no, but the early warning alert threshold was originally reached by gemini 2.5 pro and continues to be met by gemini 2.5 deep think.","publisher":"Google DeepMind","date":"August 2025","date_iso":"2025-08","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf","text":"ccl reached? no, but the early warning alert threshold was originally reached by gemini 2.5 pro and continues to be met by gemini 2.5 deep think.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0016#threshold-9eb73a502c","anchor":"threshold-9eb73a502c"},{"version":{"id":"MV-2025-0016","name":"Gemini 2.5 Deep Think","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0016"},"domain":"cyber","domain_label":"Cyber","level":"Uplift level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"for cyber uplift level 1 , the early warning alert threshold was originally reached by gemini 2.5 pro and continues to be met by gemini 2.5 deep think.","publisher":"Google DeepMind","date":"August 2025","date_iso":"2025-08","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf","text":"for cyber uplift level 1 , the early warning alert threshold was originally reached by gemini 2.5 pro and continues to be met by gemini 2.5 deep think.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0016#threshold-6e73c3fdac","anchor":"threshold-6e73c3fdac"},{"version":{"id":"MV-2025-0019","name":"Gemini 2.5 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"27 June 2025","url":"/safety/models/MV-2025-0019"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Machine Learning R&D Uplift Level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"given gemini 2.5 pro preview's poor median performance on re-bench relative to experts, the model likely lacks the necessary capabilities to automate or significantly uplift any significant fraction of the research process.","publisher":"Google DeepMind","date":"27 June 2025","date_iso":"2025-06-27","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf","text":"given gemini 2.5 pro preview's poor median performance on re-bench relative to experts, the model likely lacks the necessary capabilities to automate or significantly uplift any significant fraction of the research process.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0019#threshold-e0e02cc323","anchor":"threshold-e0e02cc323"},{"version":{"id":"MV-2025-0019","name":"Gemini 2.5 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"27 June 2025","url":"/safety/models/MV-2025-0019"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Instrumental Reasoning Level 1","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"instru- mental reasoning level 1 ✘ ccl not reached","publisher":"Google DeepMind","date":"27 June 2025","date_iso":"2025-06-27","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf","text":"instru- mental reasoning level 1 ✘ ccl not reached","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0019#threshold-02e22335e0","anchor":"threshold-02e22335e0"},{"version":{"id":"MV-2025-0019","name":"Gemini 2.5 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"27 June 2025","url":"/safety/models/MV-2025-0019"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Instrumental Reasoning Level 2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"instru- mental reasoning level 2 ✘ ccl not reached","publisher":"Google DeepMind","date":"27 June 2025","date_iso":"2025-06-27","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf","text":"instru- mental reasoning level 2 ✘ ccl not reached","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0019#threshold-751da4b11d","anchor":"threshold-751da4b11d"},{"version":{"id":"MV-2025-0023","name":"Claude Opus 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0023"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"for claude opus 4, our evaluations showed that, whereas we could confidently rule out asl-4 capabilities, we could not conclusively rule out asl-3 capabilities—which led to our precautionary decision to deploy it with asl-3 protections .","publisher":"Anthropic","date":"4 May 2025","date_iso":"2025-05-04","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf","text":"for claude opus 4, our evaluations showed that, whereas we could confidently rule out asl-4 capabilities, we could not conclusively rule out asl-3 capabilities—which led to our precautionary decision to deploy it with asl-3 protections .","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0023#threshold-0c91007de2","anchor":"threshold-0c91007de2"},{"version":{"id":"MV-2025-0023","name":"Claude Opus 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0023"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"for claude opus 4, our evaluations showed that, whereas we could confidently rule out asl-4 capabilities, we could not conclusively rule out asl-3 capabilities—which led to our precautionary decision to deploy it with asl-3 protections .","publisher":"Anthropic","date":"4 May 2025","date_iso":"2025-05-04","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf","text":"for claude opus 4, our evaluations showed that, whereas we could confidently rule out asl-4 capabilities, we could not conclusively rule out asl-3 capabilities—which led to our precautionary decision to deploy it with asl-3 protections .","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0023#threshold-99e3b55ca9","anchor":"threshold-99e3b55ca9"},{"version":{"id":"MV-2025-0022","name":"Claude Sonnet 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0022"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","publisher":"Anthropic","date":"4 May 2025","date_iso":"2025-05-04","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf","text":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0022#threshold-52087ecd28","anchor":"threshold-52087ecd28"},{"version":{"id":"MV-2025-0022","name":"Claude Sonnet 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0022"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","publisher":"Anthropic","date":"4 May 2025","date_iso":"2025-05-04","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf","text":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0022#threshold-14e15bb7c5","anchor":"threshold-14e15bb7c5"},{"version":{"id":"MV-2025-0024","name":"o3","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0024"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0024#threshold-b96c76e329","anchor":"threshold-b96c76e329"},{"version":{"id":"MV-2025-0024","name":"o3","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0024"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0024#threshold-465fa69dad","anchor":"threshold-465fa69dad"},{"version":{"id":"MV-2025-0024","name":"o3","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0024"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0024#threshold-ac9679fd1d","anchor":"threshold-ac9679fd1d"},{"version":{"id":"MV-2025-0025","name":"o4-mini","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0025"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0025#threshold-6a2fa201dd","anchor":"threshold-6a2fa201dd"},{"version":{"id":"MV-2025-0025","name":"o4-mini","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0025"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0025#threshold-7015fbd21a","anchor":"threshold-7015fbd21a"},{"version":{"id":"MV-2025-0025","name":"o4-mini","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0025"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0025#threshold-414d403dca","anchor":"threshold-414d403dca"},{"version":{"id":"MV-2025-0039","name":"Claude Sonnet 3.7","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"February 2025","url":"/safety/models/MV-2025-0039"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-4","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"these problems are harder to solve than the previous problems and we consider them as part of our asl-4 evaluations.","publisher":"Anthropic","date":"February 2025","date_iso":"2025-02","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf","text":"these problems are harder to solve than the previous problems and we consider them as part of our asl-4 evaluations.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0039#threshold-de8350a522","anchor":"threshold-de8350a522"},{"version":{"id":"MV-2025-0040","name":"Operator","provider":"OpenAI","provider_slug":"openai","released":"23 January 2025","released_iso":"2025-01-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0040"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Low","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we adapted these two evals for the computer-use setting and found the pre-mitigation operator model to be “low” risk in both categories, consistent with the gpt‑4o model.","publisher":"OpenAI","date":"23 January 2025","date_iso":"2025-01-23","url":"https://openai.com/index/operator-system-card/","text":"we adapted these two evals for the computer-use setting and found the pre-mitigation operator model to be “low” risk in both categories, consistent with the gpt‐4o model.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0040#threshold-7cd9943023","anchor":"threshold-7cd9943023"},{"version":{"id":"MV-2025-0040","name":"Operator","provider":"OpenAI","provider_slug":"openai","released":"23 January 2025","released_iso":"2025-01-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0040"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Low","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"this 1% qualifies as “low” risk, matching gpt‑4o’s “low” risk level, as detailed in openai’s preparedness framework (opens in a new window) 3 .","publisher":"OpenAI","date":"23 January 2025","date_iso":"2025-01-23","url":"https://openai.com/index/operator-system-card/","text":"this 1% qualifies as “low” risk, matching gpt‐4o’s “low” risk level, as detailed in openai’s preparedness framework (opens in a new window) 3 .","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0040#threshold-f70f98f3fb","anchor":"threshold-f70f98f3fb"}],"groups":[{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"rows":[{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"TCL (Tracked Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"In light of this, we assess that we cannot rule out the TCL.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"In light of this, we assess that we cannot rule out the TCL.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-e5a149a655","anchor":"threshold-e5a149a655"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"We cannot rule out that Gemini 4 Argon reaches the biological CCL.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"We cannot rule out that Gemini 4 Argon reaches the biological CCL.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-7a5b67e27c","anchor":"threshold-7a5b67e27c"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"TCL (Tracked Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"We can rule out the TCL for the chemical, radiological, and nuclear domains with reasonable confidence.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"We can rule out the TCL for the chemical, radiological, and nuclear domains with reasonable confidence.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-0e21e502cc","anchor":"threshold-0e21e502cc"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"cyber","domain_label":"Cyber","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"We cannot rule out that Gemini 4 Argon reaches the cybersecurity CCL.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"We cannot rule out that Gemini 4 Argon reaches the cybersecurity CCL.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-a261637daf","anchor":"threshold-a261637daf"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess its overall capabilities to be below the CCL threshold.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"we assess its overall capabilities to be below the CCL threshold.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-6819cd7a4f","anchor":"threshold-6819cd7a4f"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Stealth and situational awareness TCL (Tracked Capability Level)","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"While Gemini 4 Argon exhibits capabilities that meet our thresholds for stealth and situational awareness, we assess that the risk of severe harm or unconstrained loss of control remains low.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"While Gemini 4 Argon exhibits capabilities that meet our thresholds for stealth and situational awareness, we assess that the risk of severe harm or unconstrained loss of control remains low.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-18a3dca6dd","anchor":"threshold-18a3dca6dd"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"misuse_and_jailbreak","domain_label":"Misuse & jailbreak","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"Gemini 4 Argon demonstrates some ability to influence user beliefs and behaviors during one-on-one direct conversations in human behavioral studies, and may provide utility in the planning stage of a malicious influence operation but we assess its overall capabilities to be below the CCL threshold.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"Gemini 4 Argon demonstrates some ability to influence user beliefs and behaviors during one-on-one direct conversations in human behavioral studies, and may provide utility in the planning stage of a malicious influence operation but we assess its overall capabilities to be below the CCL threshold.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-8775192ed9","anchor":"threshold-8775192ed9"},{"version":{"id":"MV-2026-0241","name":"Gemini 4 Argon","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 October 2026","url":"/safety/models/MV-2026-0241"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"Gemini 4 Argon's performance remains below our CCL threshold.","publisher":"Google DeepMind","date":"13 October 2026","date_iso":"2026-10-13","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","text":"Gemini 4 Argon's performance remains below our CCL threshold.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0241#threshold-0bc94628a5","anchor":"threshold-0bc94628a5"}]},{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"rows":[{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Autonomy-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","publisher":"Anthropic","date":"7 October 2026","date_iso":"2026-10-07","url":"https://www.anthropic.com/claude-haiku-5-5-system-card","text":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0236#threshold-1f30090b25","anchor":"threshold-1f30090b25"},{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Autonomy-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","publisher":"Anthropic","date":"7 October 2026","date_iso":"2026-10-07","url":"https://www.anthropic.com/claude-haiku-5-5-system-card","text":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0236#threshold-800289c404","anchor":"threshold-800289c404"},{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","publisher":"Anthropic","date":"7 October 2026","date_iso":"2026-10-07","url":"https://www.anthropic.com/claude-haiku-5-5-system-card","text":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0236#threshold-588d3fff02","anchor":"threshold-588d3fff02"},{"version":{"id":"MV-2026-0236","name":"Claude Haiku 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"7 October 2026","url":"/safety/models/MV-2026-0236"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","publisher":"Anthropic","date":"7 October 2026","date_iso":"2026-10-07","url":"https://www.anthropic.com/claude-haiku-5-5-system-card","text":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","damaged":false,"lowered":false},"also":[],"anchor_url":"/safety/models/MV-2026-0236#threshold-df67b035be","anchor":"threshold-df67b035be"}]},{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"rows":[{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Autonomy-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.","damaged":false,"lowered":true},"also":[{"quote":"as such, we re-assess the risks from this threat model in comparison to the analysis in our most recent risk report for all of our models capable enough to meet our autonomy-1 threshold of performing autonomous, goal-directed operation with access to sensitive resources (which includes claude sonnet 5.5), although our level of concern is generally lower for non-frontier models.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"as such, we re-assess the risks from this threat model in comparison to the analysis in our most recent risk report for all of our models capable enough to meet our autonomy-1 threshold of performing autonomous, goal-directed operation with access to sensitive resources (which includes claude sonnet 5.5), although our level of concern is generally lower for non-frontier models.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0226#threshold-bb00b7c97a","anchor":"threshold-bb00b7c97a"},{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Autonomy-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0226#threshold-a8b13d9a43","anchor":"threshold-a8b13d9a43"},{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"we do treat sonnet 5.5 as meeting our cb-1 and autonomy-1 thresholds and apply the corresponding mitigations from our rsp.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0226#threshold-fe9ad9e5f5","anchor":"threshold-fe9ad9e5f5"},{"version":{"id":"MV-2026-0226","name":"Claude Sonnet 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"28 September 2026","released_iso":"2026-09-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0226"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.","publisher":"Anthropic","date":"28 September 2026","date_iso":"2026-09-28","url":"https://www.anthropic.com/claude-sonnet-5-5-system-card","text":"as such, our determination that opus 5.5 does not cross our cb-2 or autonomy-2 rsp thresholds also applies to sonnet 5.5.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0226#threshold-5c0ef007b8","anchor":"threshold-5c0ef007b8"}]},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"rows":[{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model: (1) we do not observe a sustained ai-attributable 2× acceleration in the pace of our overall ai progress, and (2) the model is not close to substituting for our research scientists and research engineers, especially relatively senior ones.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model: (1) we do not observe a sustained ai-attributable 2× acceleration in the pace of our overall ai progress, and (2) the model is not close to substituting for our research scientists and research engineers, especially relatively senior ones.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-d92d3b6fe9","anchor":"threshold-d92d3b6fe9"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Dramatic acceleration of automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess that claude opus 5.5 does not cross the capability threshold for dramatic acceleration of automated ai r&d in our rsp.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"we assess that claude opus 5.5 does not cross the capability threshold for dramatic acceleration of automated ai r&d in our rsp.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-632abbddea","anchor":"threshold-632abbddea"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Next capability threshold in our RSP and FCF","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"in automated ai research and development, we assess that claude opus 5.5 does not cross the next capability threshold in our rsp and fcf.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"in automated ai research and development, we assess that claude opus 5.5 does not cross the next capability threshold in our rsp and fcf.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-d1bb9898f3","anchor":"threshold-d1bb9898f3"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"on chemical and biological risks, we treat opus 5.5 as having cb-1 capabilities (relating to the synthesis of non-novel weapons) but not cb-2 capabilities (relating to the synthesis of novel weapons).","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"on chemical and biological risks, we treat opus 5.5 as having cb-1 capabilities (relating to the synthesis of non-novel weapons) but not cb-2 capabilities (relating to the synthesis of novel weapons).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-7186b0e763","anchor":"threshold-7186b0e763"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"on chemical and biological risks, we treat opus 5.5 as having cb-1 capabilities (relating to the synthesis of non-novel weapons) but not cb-2 capabilities (relating to the synthesis of novel weapons).","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"on chemical and biological risks, we treat opus 5.5 as having cb-1 capabilities (relating to the synthesis of non-novel weapons) but not cb-2 capabilities (relating to the synthesis of novel weapons).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-0fe3b7d2de","anchor":"threshold-0fe3b7d2de"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"cyber","domain_label":"Cyber","level":"Lower category of capabilities set out in our frontier compliance framework","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"despite claude opus 5.5’s cyber capabilities, the model still falls within the lower category of capabilities set out in our frontier compliance framework (fcf), and we do not see indications that the model is capable of novel offensive capability.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","text":"despite claude opus 5.5’s cyber capabilities, the model still falls within the lower category of capabilities set out in our frontier compliance framework (fcf), and we do not see indications that the model is capable of novel offensive capability.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-112c945031","anchor":"threshold-112c945031"},{"version":{"id":"MV-2026-0004","name":"Claude Opus 5.5","provider":"Anthropic","provider_slug":"anthropic","released":"22 September 2026","released_iso":"2026-09-22","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0004"},"domain":"cyber","domain_label":"Cyber","level":"Lower category of the Frontier Compliance Framework (FCF)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"despite claude opus 5.5’s cyber capabilities, the model still falls within the lower category of capabilities set out in our frontier compliance framework (fcf), and we do not see indications that the model is capable of novel offensive capability.","publisher":"Anthropic","date":"22 September 2026","date_iso":"2026-09-22","url":"https://www.anthropic.com/claude-opus-5-5-system-card","text":"despite claude opus 5.5’s cyber capabilities, the model still falls within the lower category of capabilities set out in our frontier compliance framework (fcf), and we do not see indications that the model is capable of novel offensive capability.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0004#threshold-1382e7d589","anchor":"threshold-1382e7d589"}]},{"version":{"id":"MV-2026-0005","name":"Grok 4.7","provider":"xAI","provider_slug":"xai","released":"21 September 2026","released_iso":"2026-09-21","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0005"},"rows":[{"version":{"id":"MV-2026-0005","name":"Grok 4.7","provider":"xAI","provider_slug":"xai","released":"21 September 2026","released_iso":"2026-09-21","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0005"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"FAIF safety (Frontier AI Framework)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"grok 4.7 scores below the faif safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","publisher":"xAI","date":"21 September 2026","date_iso":"2026-09-21","url":"https://media.x.ai/v1/website/card4p7-3a96f40b.pdf","text":"grok 4.7 scores below the faif safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0005#threshold-904e31f7ee","anchor":"threshold-904e31f7ee"}]},{"version":{"id":"MV-2026-0014","name":"GPT-Image-2.5-Flare","provider":"OpenAI","provider_slug":"openai","released":"8 September 2026","released_iso":"2026-09-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0014"},"rows":[{"version":{"id":"MV-2026-0014","name":"GPT-Image-2.5-Flare","provider":"OpenAI","provider_slug":"openai","released":"8 September 2026","released_iso":"2026-09-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0014"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the results indicate that neither gpt-image-2.5-sunburst nor gpt-image- 2.5-flare cross the bio high threshold or the cyber high threshold.","publisher":"OpenAI","date":"8 September 2026","date_iso":"2026-09-08","url":"https://deploymentsafety.openai.com/chatgpt-images-2-5/chatgpt-images-2-5.pdf","text":"the results indicate that neither gpt-image-2.5-sunburst nor gpt-image- 2.5-flare cross the bio high threshold or the cyber high threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0014#threshold-1e93ade02f","anchor":"threshold-1e93ade02f"},{"version":{"id":"MV-2026-0014","name":"GPT-Image-2.5-Flare","provider":"OpenAI","provider_slug":"openai","released":"8 September 2026","released_iso":"2026-09-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0014"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the results indicate that neither gpt-image-2.5-sunburst nor gpt-image- 2.5-flare cross the bio high threshold or the cyber high threshold.","publisher":"OpenAI","date":"8 September 2026","date_iso":"2026-09-08","url":"https://deploymentsafety.openai.com/chatgpt-images-2-5/chatgpt-images-2-5.pdf","text":"the results indicate that neither gpt-image-2.5-sunburst nor gpt-image- 2.5-flare cross the bio high threshold or the cyber high threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0014#threshold-9074d9d735","anchor":"threshold-9074d9d735"}]},{"version":{"id":"MV-2026-0020","name":"GPT-6 Astra","provider":"OpenAI","provider_slug":"openai","released":"3 September 2026","released_iso":"2026-09-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0020"},"rows":[{"version":{"id":"MV-2026-0020","name":"GPT-6 Astra","provider":"OpenAI","provider_slug":"openai","released":"3 September 2026","released_iso":"2026-09-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0020"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we are treating astra as high capability in the biological and chemical domain.","publisher":"OpenAI","date":"3 September 2026","date_iso":"2026-09-03","url":"https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf","text":"we are treating astra as high capability in the biological and chemical domain.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0020#threshold-976d12867e","anchor":"threshold-976d12867e"},{"version":{"id":"MV-2026-0020","name":"GPT-6 Astra","provider":"OpenAI","provider_slug":"openai","released":"3 September 2026","released_iso":"2026-09-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0020"},"domain":"cyber","domain_label":"Cyber","level":"Critical","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"based on our evaluations, we believe that astra meets the critical cybersecurity threshold.","publisher":"OpenAI","date":"3 September 2026","date_iso":"2026-09-03","url":"https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf","text":"based on our evaluations, we believe that astra meets the critical cybersecurity threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0020#threshold-fc9263e086","anchor":"threshold-fc9263e086"}]},{"version":{"id":"MV-2026-0025","name":"Claude Mythos 5.1","provider":"Anthropic","provider_slug":"anthropic","released":"1 September 2026","released_iso":"2026-09-01","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0025"},"rows":[{"version":{"id":"MV-2026-0025","name":"Claude Mythos 5.1","provider":"Anthropic","provider_slug":"anthropic","released":"1 September 2026","released_iso":"2026-09-01","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0025"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Capability threshold for dramatic acceleration of automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess that claude mythos 5.1 does not cross the capability threshold for dramatic acceleration of automated ai r&d in our rsp and fcf, although we think it provides a meaningful acceleration to ai r&d efforts in many circumstances, and is somewhat more capable in this domain than the models described in our previous system cards.","publisher":"Anthropic","date":"1 September 2026","date_iso":"2026-09-01","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card","text":"we assess that claude mythos 5.1 does not cross the capability threshold for dramatic acceleration of automated ai r&d in our rsp and fcf, although we think it provides a meaningful acceleration to ai r&d efforts in many circumstances, and is somewhat more capable in this domain than the models described in our previous system cards.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0025#threshold-2dad838ed9","anchor":"threshold-2dad838ed9"},{"version":{"id":"MV-2026-0025","name":"Claude Mythos 5.1","provider":"Anthropic","provider_slug":"anthropic","released":"1 September 2026","released_iso":"2026-09-01","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0025"},"domain":"cyber","domain_label":"Cyber","level":"TIER 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"although mythos 5.1 is in tier 1, it is getting closer to tier 2, completing more and more autonomous tasks.","publisher":"Anthropic","date":"1 September 2026","date_iso":"2026-09-01","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card","text":"although mythos 5.1 is in tier 1, it is getting closer to tier 2, completing more and more autonomous tasks.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0025#threshold-ee41cdfe3a","anchor":"threshold-ee41cdfe3a"}]},{"version":{"id":"MV-2026-0033","name":"Gemini 3.5 Audio","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 August 2026","url":"/safety/models/MV-2026-0033"},"rows":[{"version":{"id":"MV-2026-0033","name":"Gemini 3.5 Audio","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 August 2026","url":"/safety/models/MV-2026-0033"},"domain":"other","domain_label":"Other","level":"Tracked or critical capability level (t/ccls)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"they did not reach any tracked or critical capability levels (t/ccls) outlined in our frontier safety framework .","publisher":"Google DeepMind","date":"26 August 2026","date_iso":"2026-08-26","url":"https://deepmind.google/models/model-cards/gemini-3-5-audio/","text":"they did not reach any tracked or critical capability levels (t/ccls) outlined in our frontier safety framework .","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0033#threshold-fadaf5631d","anchor":"threshold-fadaf5631d"}]},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"rows":[{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the model does not reach the ccl alert threshold.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"the model does not reach the ccl alert threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-e3c65e47b1","anchor":"threshold-e3c65e47b1"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we can rule out the ccl for the cbrn domain with reasonable confidence based on the results from our testing.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"we can rule out the ccl for the cbrn domain with reasonable confidence based on the results from our testing.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-aac36c2a6d","anchor":"threshold-aac36c2a6d"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"TCL (Tracked Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we can rule out the tcl for the cbrn domain with reasonable confidence based on the results from our testing.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"we can rule out the tcl for the cbrn domain with reasonable confidence based on the results from our testing.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-4d71a24b8a","anchor":"threshold-4d71a24b8a"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"cyber","domain_label":"Cyber","level":"CCL (Critical Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"gemini 3.7 flash reaches the alert threshold for this ccl, but not the ccl.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"gemini 3.7 flash reaches the alert threshold for this ccl, but not the ccl.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-36b5e8b6f2","anchor":"threshold-36b5e8b6f2"},{"version":{"id":"MV-2026-0046","name":"Gemini 3.7 Flash","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"13 August 2026","url":"/safety/models/MV-2026-0046"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"TCL (Tracked Capability Level)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the model does not reach the tcl.","publisher":"Google DeepMind","date":"13 August 2026","date_iso":"2026-08-13","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf","text":"the model does not reach the tcl.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0046#threshold-4f1ae689a7","anchor":"threshold-4f1ae689a7"}]},{"version":{"id":"MV-2026-0048","name":"Grok 4.6","provider":"xAI","provider_slug":"xai","released":"12 August 2026","released_iso":"2026-08-12","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0048"},"rows":[{"version":{"id":"MV-2026-0048","name":"Grok 4.6","provider":"xAI","provider_slug":"xai","released":"12 August 2026","released_iso":"2026-08-12","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0048"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"FAIF safety (Frontier AI Framework)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"grok 4.6 scores below the faif safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","publisher":"xAI","date":"12 August 2026","date_iso":"2026-08-12","url":"https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf","text":"grok 4.6 scores below the faif safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0048#threshold-ebbd582fcd","anchor":"threshold-ebbd582fcd"}]},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"rows":[{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-2534ddcfa9","anchor":"threshold-2534ddcfa9"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower risk (inferred)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"- loss of control: moderate or lower risk (inferred).","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"- loss of control: moderate or lower risk (inferred).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-6ac2e0eeed","anchor":"threshold-6ac2e0eeed"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-20c7926cb4","anchor":"threshold-20c7926cb4"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"chem/bio: moderate or lower risk; cyber: moderate or lower risk (inferred); loss of control: moderate or lower risk (inferred).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-a4f8c43c06","anchor":"threshold-a4f8c43c06"},{"version":{"id":"MV-2026-0052","name":"Muse Glimmer 30B","provider":"Meta","provider_slug":"meta","released":"August 2026","released_iso":"2026-08","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0052"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower risk (inferred)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"- cyber: moderate or lower risk (inferred);","publisher":"Meta","date":"August 2026","date_iso":"2026-08","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B","text":"- cyber: moderate or lower risk (inferred);","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0052#threshold-2b51266ba1","anchor":"threshold-2b51266ba1"}]},{"version":{"id":"MV-2026-0071","name":"Claude Opus 5","provider":"Anthropic","provider_slug":"anthropic","released":"24 July 2026","released_iso":"2026-07-24","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0071"},"rows":[{"version":{"id":"MV-2026-0071","name":"Claude Opus 5","provider":"Anthropic","provider_slug":"anthropic","released":"24 July 2026","released_iso":"2026-07-24","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0071"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"on chemical and biological risks, we treat the model as having cb-1 capabilities (relating to the synthesis of non-novel weapons), but not cb-2 capabilities (relating to the synthesis of novel weapons).","publisher":"Anthropic","date":"24 July 2026","date_iso":"2026-07-24","url":"https://www-cdn.anthropic.com/b514064af1408018e64b1ad24e7d5e75850b4ffd/Claude%20Opus%205%20System%20Card.pdf","text":"on chemical and biological risks, we treat the model as having cb-1 capabilities (relating to the synthesis of non-novel weapons), but not cb-2 capabilities (relating to the synthesis of novel weapons).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0071#threshold-e902771bf0","anchor":"threshold-e902771bf0"},{"version":{"id":"MV-2026-0071","name":"Claude Opus 5","provider":"Anthropic","provider_slug":"anthropic","released":"24 July 2026","released_iso":"2026-07-24","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0071"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"on chemical and biological risks, we treat the model as having cb-1 capabilities (relating to the synthesis of non-novel weapons), but not cb-2 capabilities (relating to the synthesis of novel weapons).","publisher":"Anthropic","date":"24 July 2026","date_iso":"2026-07-24","url":"https://www-cdn.anthropic.com/b514064af1408018e64b1ad24e7d5e75850b4ffd/Claude%20Opus%205%20System%20Card.pdf","text":"on chemical and biological risks, we treat the model as having cb-1 capabilities (relating to the synthesis of non-novel weapons), but not cb-2 capabilities (relating to the synthesis of novel weapons).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0071#threshold-0ea0c6ea70","anchor":"threshold-0ea0c6ea70"}]},{"version":{"id":"MV-2026-0075","name":"Gemini 3.5 Flash-Lite","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"21 July 2026","url":"/safety/models/MV-2026-0075"},"rows":[{"version":{"id":"MV-2026-0075","name":"Gemini 3.5 Flash-Lite","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"21 July 2026","url":"/safety/models/MV-2026-0075"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that, while gemini 3.5 flash-lite excels at agents and coding, it does not have meaningful new capabilities or material increases in performance with respect to the domains outlined in our frontier safety framework compared to gemini 3.1 pro; therefore, based on gemini 3.1 pro results, we are confident that gemini 3.5 flash-lite is also unlikely to reach any ccls.","publisher":"Google DeepMind","date":"21 July 2026","date_iso":"2026-07-21","url":"https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/","text":"our assessments have shown that, while gemini 3.5 flash-lite excels at agents and coding, it does not have meaningful new capabilities or material increases in performance with respect to the domains outlined in our frontier safety framework compared to gemini 3.1 pro; therefore, based on gemini 3.1 pro results, we are confident that gemini 3.5 flash-lite is also unlikely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0075#threshold-384b2b75e1","anchor":"threshold-384b2b75e1"}]},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"rows":[{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Capability checkpoint","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"muse spark 1.1 demonstrates improved coding capabilities, which are now meeting one of the indicative thresholds introduced in muse spark safety & preparedness report (specifically, resolving over half of unique tasks on swe-bench verified hard).","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"muse spark 1.1 demonstrates improved coding capabilities, which are now meeting one of the indicative thresholds introduced in muse spark safety & preparedness report (specifically, resolving over half of unique tasks on swe-bench verified hard).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-ae18f623b6","anchor":"threshold-ae18f623b6"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"in the loss of control domain, the model’s capabilities remain within the “moderate or lower” threshold under meta’s advanced ai scaling framework.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"in the loss of control domain, the model’s capabilities remain within the “moderate or lower” threshold under meta’s advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-1b3a6516f6","anchor":"threshold-1b3a6516f6"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments indicate that, when evaluated without mitigations applied, we cannot rule out muse spark 1.1’s capabilities meeting this threshold in both the chemical & biological and cybersecurity domains.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"our assessments indicate that, when evaluated without mitigations applied, we cannot rule out muse spark 1.1’s capabilities meeting this threshold in both the chemical & biological and cybersecurity domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-505fbbce0d","anchor":"threshold-505fbbce0d"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments indicate that, when evaluated without mitigations applied, we cannot rule out muse spark 1.1’s capabilities meeting this threshold in both the chemical & biological and cybersecurity domains.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"our assessments indicate that, when evaluated without mitigations applied, we cannot rule out muse spark 1.1’s capabilities meeting this threshold in both the chemical & biological and cybersecurity domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-894f659c86","anchor":"threshold-894f659c86"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"therefore, we conclude that we have addressed these risks from unmitigated muse spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under advanced ai scaling framework.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"therefore, we conclude that we have addressed these risks from unmitigated muse spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-6b5cfc2380","anchor":"threshold-6b5cfc2380"},{"version":{"id":"MV-2026-0089","name":"Muse Spark 1.1","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 July 2026","url":"/safety/models/MV-2026-0089"},"domain":"misuse_and_jailbreak","domain_label":"Misuse & jailbreak","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"therefore, we conclude that we have addressed these risks from unmitigated muse spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under our advanced ai scaling framework.","publisher":"Meta","date":"9 July 2026","date_iso":"2026-07-09","url":"https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report","text":"therefore, we conclude that we have addressed these risks from unmitigated muse spark 1.1, which bring the residual risk to the “moderate or lower” risk threshold under our advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0089#threshold-ddc93b87b7","anchor":"threshold-ddc93b87b7"}]},{"version":{"id":"MV-2026-0099","name":"Claude Sonnet 5","provider":"Anthropic","provider_slug":"anthropic","released":"30 June 2026","released_iso":"2026-06-30","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0099"},"rows":[{"version":{"id":"MV-2026-0099","name":"Claude Sonnet 5","provider":"Anthropic","provider_slug":"anthropic","released":"30 June 2026","released_iso":"2026-06-30","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0099"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"sonnet 5 does not cross the cb-2 threshold.","publisher":"Anthropic","date":"30 June 2026","date_iso":"2026-06-30","url":"https://www-cdn.anthropic.com/9e6a1044980d8c4ed85669faf9c2a8342e2e9f1e/Claude%20Sonnet%205%20System%20Card.pdf","text":"sonnet 5 does not cross the cb-2 threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0099#threshold-d8f02bfc4c","anchor":"threshold-d8f02bfc4c"},{"version":{"id":"MV-2026-0099","name":"Claude Sonnet 5","provider":"Anthropic","provider_slug":"anthropic","released":"30 June 2026","released_iso":"2026-06-30","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0099"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Notable capability","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"we consider performance on the virology tasks notable if: ● the model scores above 0.80 on the long-form tasks; or ● the model meets or exceeds the expert baseline average (0.221) on the multimodal task.","publisher":"Anthropic","date":"30 June 2026","date_iso":"2026-06-30","url":"https://www-cdn.anthropic.com/9e6a1044980d8c4ed85669faf9c2a8342e2e9f1e/Claude%20Sonnet%205%20System%20Card.pdf","text":"we consider performance on the virology tasks notable if: ● the model scores above 0.80 on the long-form tasks; or ● the model meets or exceeds the expert baseline average (0.221) on the multimodal task.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0099#threshold-c24df75a69","anchor":"threshold-c24df75a69"}]},{"version":{"id":"MV-2026-0100","name":"Claude Mythos 5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"9 June 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0100"},"rows":[{"version":{"id":"MV-2026-0100","name":"Claude Mythos 5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"9 June 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0100"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model, mythos preview.","publisher":"Anthropic","date":"9 June 2026","date_iso":"2026-06-09","url":"https://www.anthropic.com/claude-fable-5-mythos-5-system-card","text":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model, mythos preview.","damaged":false,"lowered":true},"also":[{"quote":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model, mythos preview.","publisher":"","date":"9 June 2026","date_iso":"2026-06-09","url":"https://anthropic.com/claude-fable-5-mythos-5-system-card","text":"we conclude the risk threshold is not crossed, on the same two grounds as our determination for our previous frontier model, mythos preview.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0100#threshold-dee73e4d13","anchor":"threshold-dee73e4d13"}]},{"version":{"id":"MV-2026-0109","name":"Gemini 3.5 Transcribe","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 June 2026","url":"/safety/models/MV-2026-0109"},"rows":[{"version":{"id":"MV-2026-0109","name":"Gemini 3.5 Transcribe","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 June 2026","url":"/safety/models/MV-2026-0109"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that none of gemini 3.5 translate, gemini 3.5 transcribe, or gemini 3.5 transcribe live have meaningful new t/ccl-relevant capabilities or material increases in performance with respect to frontier safety compared to gemini 3.1 pro and gemini 3.7 flash, therefore we are confident that none of them are likely to reach any ccls.","publisher":"Google DeepMind","date":"9 June 2026","date_iso":"2026-06-09","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-5-Audio-Model-Card.pdf","text":"our assessments have shown that none of gemini 3.5 translate, gemini 3.5 transcribe, or gemini 3.5 transcribe live have meaningful new t/ccl-relevant capabilities or material increases in performance with respect to frontier safety compared to gemini 3.1 pro and gemini 3.7 flash, therefore we are confident that none of them are likely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0109#threshold-49960a3c41","anchor":"threshold-49960a3c41"}]},{"version":{"id":"MV-2026-0110","name":"Gemini 3.5 Transcribe Live","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 June 2026","url":"/safety/models/MV-2026-0110"},"rows":[{"version":{"id":"MV-2026-0110","name":"Gemini 3.5 Transcribe Live","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"9 June 2026","url":"/safety/models/MV-2026-0110"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that none of gemini 3.5 translate, gemini 3.5 transcribe, or gemini 3.5 transcribe live have meaningful new t/ccl-relevant capabilities or material increases in performance with respect to frontier safety compared to gemini 3.1 pro and gemini 3.7 flash, therefore we are confident that none of them are likely to reach any ccls.","publisher":"Google DeepMind","date":"9 June 2026","date_iso":"2026-06-09","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-5-Audio-Model-Card.pdf","text":"our assessments have shown that none of gemini 3.5 translate, gemini 3.5 transcribe, or gemini 3.5 transcribe live have meaningful new t/ccl-relevant capabilities or material increases in performance with respect to frontier safety compared to gemini 3.1 pro and gemini 3.7 flash, therefore we are confident that none of them are likely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0110#threshold-756533ff5d","anchor":"threshold-756533ff5d"}]},{"version":{"id":"MV-2026-0114","name":"GPT-Rosalind-5.5","provider":"OpenAI","provider_slug":"openai","released":"3 June 2026","released_iso":"2026-06-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0114"},"rows":[{"version":{"id":"MV-2026-0114","name":"GPT-Rosalind-5.5","provider":"OpenAI","provider_slug":"openai","released":"3 June 2026","released_iso":"2026-06-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0114"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"because gpt- rosalind-5.5 was trained on additional biology data unrelated to ai self-improvement, we rely on gpt-5.5’s existing preparedness assessment for ai self-improvement, where gpt-5.5 did not meet the threshold for high capability.","publisher":"OpenAI","date":"3 June 2026","date_iso":"2026-06-03","url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","text":"because gpt- rosalind-5.5 was trained on additional biology data unrelated to ai self-improvement, we rely on gpt-5.5’s existing preparedness assessment for ai self-improvement, where gpt-5.5 did not meet the threshold for high capability.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0114#threshold-5bd390c4a0","anchor":"threshold-5bd390c4a0"},{"version":{"id":"MV-2026-0114","name":"GPT-Rosalind-5.5","provider":"OpenAI","provider_slug":"openai","released":"3 June 2026","released_iso":"2026-06-03","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0114"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"under our preparedness framework, the preparedness evaluations in the biological and chemical domain met our threshold for high capability while falling below the threshold for critical.","publisher":"OpenAI","date":"3 June 2026","date_iso":"2026-06-03","url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","text":"under our preparedness framework, the preparedness evaluations in the biological and chemical domain met our threshold for high capability while falling below the threshold for critical.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0114#threshold-1445fc72e4","anchor":"threshold-1445fc72e4"}]},{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"rows":[{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"loss of control risks.we assess loss of control risk as “moderate or lower.”","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"loss of control risks.we assess loss of control risk as “moderate or lower.”","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0115#threshold-34d87b7657","anchor":"threshold-34d87b7657"},{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we retain the same “high risk” determination for the unmitigated model.","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"we retain the same “high risk” determination for the unmitigated model.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0115#threshold-0e70889ea3","anchor":"threshold-0e70889ea3"},{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"after validating that the mitigations deployed for muse spark, including refusal mechanisms, continue to perform robustly under the contemplating architecture, we confirm residual risk at “moderate or lower.”","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"after validating that the mitigations deployed for muse spark, including refusal mechanisms, continue to perform robustly under the contemplating architecture, we confirm residual risk at “moderate or lower.”","damaged":false,"lowered":true},"also":[{"quote":"after applying the same mitigations as for muse spark, the residual risk is reduced to “moderate or lower.”","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"after applying the same mitigations as for muse spark, the residual risk is reduced to “moderate or lower.”","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0115#threshold-c251b7dd0f","anchor":"threshold-c251b7dd0f"},{"version":{"id":"MV-2026-0115","name":"Muse Spark Contemplating","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 June 2026","url":"/safety/models/MV-2026-0115"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in meta’s advanced ai scaling framework.","publisher":"Meta","date":"3 June 2026","date_iso":"2026-06-03","url":"https://ai.meta.com/static-resource/muse-spark-contemplating-safety-and-preparedness-report/","text":"we assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in meta’s advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0115#threshold-1696252301","anchor":"threshold-1696252301"}]},{"version":{"id":"MV-2026-0098","name":"Gemini 3.1 Flash-Lite Image","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"June 2026","released_iso":"2026-06","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0098"},"rows":[{"version":{"id":"MV-2026-0098","name":"Gemini 3.1 Flash-Lite Image","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"June 2026","released_iso":"2026-06","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0098"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that gemini 3.1 flash-lite image is less capable than gemini 3.1 pro, therefore based on gemini 3.1 pro, we are confident that that gemini 3.1 flash-lite image is also unlikely to reach any ccls.","publisher":"Google DeepMind","date":"June 2026","date_iso":"2026-06","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Flash-Lite-Image-Model-Card.pdf","text":"our assessments have shown that gemini 3.1 flash-lite image is less capable than gemini 3.1 pro, therefore based on gemini 3.1 pro, we are confident that that gemini 3.1 flash-lite image is also unlikely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0098#threshold-4fd12b5ca1","anchor":"threshold-4fd12b5ca1"}]},{"version":{"id":"MV-2026-0125","name":"Claude Opus 4.8","provider":"Anthropic","provider_slug":"anthropic","released":"28 May 2026","released_iso":"2026-05-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0125"},"rows":[{"version":{"id":"MV-2026-0125","name":"Claude Opus 4.8","provider":"Anthropic","provider_slug":"anthropic","released":"28 May 2026","released_iso":"2026-05-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0125"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess that claude opus 4.8 does not cross the automated ai-r&d capability threshold.","publisher":"Anthropic","date":"28 May 2026","date_iso":"2026-05-28","url":"https://www.anthropic.com/claude-opus-4-8-system-card","text":"we assess that claude opus 4.8 does not cross the automated ai-r&d capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0125#threshold-c127a687aa","anchor":"threshold-c127a687aa"},{"version":{"id":"MV-2026-0125","name":"Claude Opus 4.8","provider":"Anthropic","provider_slug":"anthropic","released":"28 May 2026","released_iso":"2026-05-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0125"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"non-novel chemical and biological weapons (cb-1): we believe it is hard to be confident whether a model passes this threshold, but (as with previous recent models) our results are consistent with opus 4.8 providing significant uplift to individuals or groups with basic technical backgrounds for this threat model.","publisher":"Anthropic","date":"28 May 2026","date_iso":"2026-05-28","url":"https://www.anthropic.com/claude-opus-4-8-system-card","text":"non-novel chemical and biological weapons (cb-1): we believe it is hard to be confident whether a model passes this threshold, but (as with previous recent models) our results are consistent with opus 4.8 providing significant uplift to individuals or groups with basic technical backgrounds for this threat model.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0125#threshold-3d542dcc6a","anchor":"threshold-3d542dcc6a"},{"version":{"id":"MV-2026-0125","name":"Claude Opus 4.8","provider":"Anthropic","provider_slug":"anthropic","released":"28 May 2026","released_iso":"2026-05-28","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0125"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"because opus 4.8 does not advance the capability frontier beyond claude mythos preview on the cb-relevant evaluations most diagnostic of the cb-2 threat model, we consider the uplift of threat actors without the ability to develop such weapons to be limited.","publisher":"Anthropic","date":"28 May 2026","date_iso":"2026-05-28","url":"https://www.anthropic.com/claude-opus-4-8-system-card","text":"because opus 4.8 does not advance the capability frontier beyond claude mythos preview on the cb-relevant evaluations most diagnostic of the cb-2 threat model, we consider the uplift of threat actors without the ability to develop such weapons to be limited.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0125#threshold-e5a6c69ce6","anchor":"threshold-e5a6c69ce6"}]},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"rows":[{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess loss of control risk as “moderate or lower”.","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"we assess loss of control risk as “moderate or lower”.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-f1da2a69f4","anchor":"threshold-f1da2a69f4"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"those evaluations identified elevated risks prior to mitigations, with chemical and biological capabilities assessed as likely reaching the “high risk” category under the advanced ai scaling framework before safeguards were applied.","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"those evaluations identified elevated risks prior to mitigations, with chemical and biological capabilities assessed as likely reaching the “high risk” category under the advanced ai scaling framework before safeguards were applied.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-4c6c2a4863","anchor":"threshold-4c6c2a4863"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"however, after applying mitigations, the residual risk of the model is that of “moderate or lower risk.”","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"however, after applying mitigations, the residual risk of the model is that of “moderate or lower risk.”","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-d040130bc3","anchor":"threshold-d040130bc3"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"cyber","domain_label":"Cyber","level":"Below the level required to meaningfully enable the cyber threat scenarios defin","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"muse spark’s offensive cyber capabilities remain below the level required to meaningfully enable the cyber threat scenarios defined in the advanced ai scaling framework.","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"muse spark’s offensive cyber capabilities remain below the level required to meaningfully enable the cyber threat scenarios defined in the advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-3bb6399c37","anchor":"threshold-3bb6399c37"},{"version":{"id":"MV-2026-0129","name":"Muse Spark","provider":"Meta","provider_slug":"meta","released":"","released_iso":"","card_dated":"","provider_doc_dated":"26 May 2026","url":"/safety/models/MV-2026-0129"},"domain":"cyber","domain_label":"Cyber","level":"Moderate or lower","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in meta’s advanced ai scaling framework.","publisher":"Meta","date":"26 May 2026","date_iso":"2026-05-26","url":"https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/","text":"we assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in meta’s advanced ai scaling framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0129#threshold-19a0b4617e","anchor":"threshold-19a0b4617e"}]},{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"rows":[{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Below high capability","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"gpt-5.5 instant is considered below high capability.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"gpt-5.5 instant is considered below high capability.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0135#threshold-5b6a76027d","anchor":"threshold-5b6a76027d"},{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.5 instant does not have a plausible chance of reaching a high threshold.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.5 instant does not have a plausible chance of reaching a high threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0135#threshold-78bb6c1440","anchor":"threshold-78bb6c1440"},{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"gpt-5.5 instant is our first instant model to be treated as high capability in the biological and chemical domain, as well as in the cybersecurity domain.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"gpt-5.5 instant is our first instant model to be treated as high capability in the biological and chemical domain, as well as in the cybersecurity domain.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0135#threshold-b623243f6e","anchor":"threshold-b623243f6e"},{"version":{"id":"MV-2026-0135","name":"GPT-5.5 Instant","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"4 May 2026","provider_doc_dated":"","url":"/safety/models/MV-2026-0135"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"this is the first instant model that we are treating as high capability in our cybersecurity and biological & chemical preparedness categories, and implementing appropriate safeguards.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"this is the first instant model that we are treating as high capability in our cybersecurity and biological & chemical preparedness categories, and implementing appropriate safeguards.","damaged":false,"lowered":true},"also":[{"quote":"because of this, we similarly treat gpt-5.5 instant as high under the preparedness framework.","publisher":"OpenAI","date":"4 May 2026","date_iso":"2026-05-04","url":"https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf","text":"because of this, we similarly treat gpt-5.5 instant as high under the preparedness framework.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0135#threshold-06d06fe664","anchor":"threshold-06d06fe664"}]},{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"rows":[{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.5 does not have a plausible chance of reaching a high threshold.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.5 does not have a plausible chance of reaching a high threshold.","damaged":false,"lowered":true},"also":[{"quote":"gpt-5.5 did not meet our thresholds for high capability in ai self-improvement.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"gpt-5.5 did not meet our thresholds for high capability in ai self-improvement.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0154#threshold-cfe2b3524c","anchor":"threshold-cfe2b3524c"},{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we are continuing to treat gpt-5.5 as high capability in the biological and chemical domain.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"we are continuing to treat gpt-5.5 as high capability in the biological and chemical domain.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0154#threshold-5ea957e9c3","anchor":"threshold-5ea957e9c3"},{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"domain":"cyber","domain_label":"Cyber","level":"Critical","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we therefore treat vulnlmp as important evidence that gpt-5.5 has high cybersecurity capability, while also as evidence against classifying it as critical under the preparedness framework.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"we therefore treat vulnlmp as important evidence that gpt-5.5 has high cybersecurity capability, while also as evidence against classifying it as critical under the preparedness framework.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0154#threshold-3d972a74f4","anchor":"threshold-3d972a74f4"},{"version":{"id":"MV-2026-0154","name":"GPT-5.5","provider":"OpenAI","provider_slug":"openai","released":"23 April 2026","released_iso":"2026-04-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0154"},"domain":"cyber","domain_label":"Cyber","level":"High capability, below critical","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we are treating gpt-5.5 as high capability in the cybersecurity domain, but below critical.","publisher":"OpenAI","date":"23 April 2026","date_iso":"2026-04-23","url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","text":"we are treating gpt-5.5 as high capability in the cybersecurity domain, but below critical.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0154#threshold-ad63b86a7d","anchor":"threshold-ad63b86a7d"}]},{"version":{"id":"MV-2026-0167","name":"Claude Opus 4.7","provider":"Anthropic","provider_slug":"anthropic","released":"16 April 2026","released_iso":"2026-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0167"},"rows":[{"version":{"id":"MV-2026-0167","name":"Claude Opus 4.7","provider":"Anthropic","provider_slug":"anthropic","released":"16 April 2026","released_iso":"2026-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0167"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"that opus 4.7 does not cross the threshold for automated ai r&d;","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://www.anthropic.com/claude-opus-4-7-system-card","text":"that opus 4.7 does not cross the threshold for automated ai r&d;","damaged":false,"lowered":true},"also":[{"quote":"we assess that claude opus 4.7 does not cross the automated ai-r&d capability threshold.","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://www.anthropic.com/claude-opus-4-7-system-card","text":"we assess that claude opus 4.7 does not cross the automated ai-r&d capability threshold.","damaged":false,"lowered":true},{"quote":"that opus 4.7 does not cross the threshold for automated ai r&d; and that misalignment risk remains very low (though higher than for pre-mythos preview models).","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://anthropic.com/claude-opus-4-7-system-card","text":"that opus 4.7 does not cross the threshold for automated ai r&d; and that misalignment risk remains very low (though higher than for pre-mythos preview models).","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0167#threshold-1d015e9fdb","anchor":"threshold-1d015e9fdb"},{"version":{"id":"MV-2026-0167","name":"Claude Opus 4.7","provider":"Anthropic","provider_slug":"anthropic","released":"16 April 2026","released_iso":"2026-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0167"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"we believe it is hard to be confident regarding whether a model passes this threshold.","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://www.anthropic.com/claude-opus-4-7-system-card","text":"we believe it is hard to be confident regarding whether a model passes this threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0167#threshold-09307b504a","anchor":"threshold-09307b504a"},{"version":{"id":"MV-2026-0167","name":"Claude Opus 4.7","provider":"Anthropic","provider_slug":"anthropic","released":"16 April 2026","released_iso":"2026-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0167"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our evaluations suggest that the model is not yet at the level of capability associated with the cb-2 threat model (above).","publisher":"Anthropic","date":"16 April 2026","date_iso":"2026-04-16","url":"https://www.anthropic.com/claude-opus-4-7-system-card","text":"our evaluations suggest that the model is not yet at the level of capability associated with the cb-2 threat model (above).","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0167#threshold-8fe8052e3a","anchor":"threshold-8fe8052e3a"}]},{"version":{"id":"MV-2026-0170","name":"Gemini 3.1 Flash Audio","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"15 April 2026","url":"/safety/models/MV-2026-0170"},"rows":[{"version":{"id":"MV-2026-0170","name":"Gemini 3.1 Flash Audio","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"15 April 2026","url":"/safety/models/MV-2026-0170"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that gemini 3.1 flash audio is less capable than gemini 3.1 pro.","publisher":"Google DeepMind","date":"15 April 2026","date_iso":"2026-04-15","url":"https://deepmind.google/models/model-cards/gemini-3-1-flash-audio/","text":"our assessments have shown that gemini 3.1 flash audio is less capable than gemini 3.1 pro.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0170#threshold-7829dbbc8f","anchor":"threshold-7829dbbc8f"}]},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"rows":[{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D rule-out","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the main reason we have determined that claude mythos preview does not cross the threshold in question is that we have been using it extensively in the course of our day-to-day work and exploring where it can automate such work, and it does not seem close to being able to substitute for research scientists and research engineers—especially relatively senior ones.","publisher":"","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www.anthropic.com/claude-mythos-preview-system-card","text":"the main reason we have determined that claude mythos preview does not cross the threshold in question is that we have been using it extensively in the course of our day-to-day work and exploring where it can automate such work, and it does not seem close to being able to substitute for research scientists and research engineers—especially relatively senior ones.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0181#threshold-78837e5e82","anchor":"threshold-78837e5e82"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we assess that claude mythos preview does not cross the automated ai-r&d capability threshold.","publisher":"Anthropic","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","text":"we assess that claude mythos preview does not cross the automated ai-r&d capability threshold.","damaged":false,"lowered":true},"also":[{"quote":"we assess that claude mythos preview does not cross the automated ai-r&d capability threshold.","publisher":"","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www.anthropic.com/claude-mythos-preview-system-card","text":"we assess that claude mythos preview does not cross the automated ai-r&d capability threshold.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0181#threshold-5c424f12f2","anchor":"threshold-5c424f12f2"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"RSP automated AI R&D","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we have concluded that claude mythos preview does not cross the rsp automated ai r&d threshold of compressing two years of progress into one.","publisher":"Anthropic","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","text":"we have concluded that claude mythos preview does not cross the rsp automated ai r&d threshold of compressing two years of progress into one.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0181#threshold-fbcd7989d6","anchor":"threshold-fbcd7989d6"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"we believe it is hard to be confident regarding whether a model passes this threshold.","publisher":"Anthropic","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","text":"we believe it is hard to be confident regarding whether a model passes this threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0181#threshold-70183ff1fb","anchor":"threshold-70183ff1fb"},{"version":{"id":"MV-2026-0181","name":"Claude Mythos Preview","provider":"Anthropic","provider_slug":"anthropic","released":"7 April 2026","released_iso":"2026-04-07","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0181"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CB-2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we believe that claude mythos preview does not pass this threshold due to its noted limitations in open-ended scientific reasoning, strategic judgment, and hypothesis triage.","publisher":"Anthropic","date":"7 April 2026","date_iso":"2026-04-07","url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","text":"we believe that claude mythos preview does not pass this threshold due to its noted limitations in open-ended scientific reasoning, strategic judgment, and hypothesis triage.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0181#threshold-a224925ca4","anchor":"threshold-a224925ca4"}]},{"version":{"id":"MV-2026-0200","name":"GPT-5.4 Thinking","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0200"},"rows":[{"version":{"id":"MV-2026-0200","name":"GPT-5.4 Thinking","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0200"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.4-thinking does not have a plausible chance of reaching a high threshold.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"for ai self-improvement, evaluations of final checkpoints indicate that, like its predecessor models, gpt-5.4-thinking does not have a plausible chance of reaching a high threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0200#threshold-69cf502d31","anchor":"threshold-69cf502d31"},{"version":{"id":"MV-2026-0200","name":"GPT-5.4 Thinking","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0200"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"as we did for gpt-5.2-thinking before it, we are continuing to treat gpt-5.4-thinking as high capability in the biological and chemical domain.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"as we did for gpt-5.2-thinking before it, we are continuing to treat gpt-5.4-thinking as high capability in the biological and chemical domain.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0200#threshold-5fc91c31d0","anchor":"threshold-5fc91c31d0"},{"version":{"id":"MV-2026-0200","name":"GPT-5.4 Thinking","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0200"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"as we did for gpt-5.3-codex, we are treating gpt-5.4-thinking as high capability in the cybersecurity domain, and applied safeguards as described in the safeguard section below.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"as we did for gpt-5.3-codex, we are treating gpt-5.4-thinking as high capability in the cybersecurity domain, and applied safeguards as described in the safeguard section below.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0200#threshold-a12e61e384","anchor":"threshold-a12e61e384"}]},{"version":{"id":"MV-2026-0201","name":"GPT-5.4 mini","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0201"},"rows":[{"version":{"id":"MV-2026-0201","name":"GPT-5.4 mini","provider":"OpenAI","provider_slug":"openai","released":"5 March 2026","released_iso":"2026-03-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0201"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"gpt-5.4 mini’s biology capabilities are closely comparable to those of gpt-5. we conclude that it, too, is not high under our preparedness framework.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"gpt-5.4 mini’s biology capabilities are closely comparable to those of gpt-5. we conclude that it, too, is not high under our preparedness framework.","damaged":false,"lowered":true},"also":[{"quote":"we therefore conclude that gpt-5.4 mini should not be classified as bio high under the current preparedness framework novice-uplift criterion.","publisher":"OpenAI","date":"5 March 2026","date_iso":"2026-03-05","url":"https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf","text":"we therefore conclude that gpt-5.4 mini should not be classified as bio high under the current preparedness framework novice-uplift criterion.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2026-0201#threshold-762f00523f","anchor":"threshold-762f00523f"}]},{"version":{"id":"MV-2026-0203","name":"Gemini 3.1 Flash-Lite","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 March 2026","url":"/safety/models/MV-2026-0203"},"rows":[{"version":{"id":"MV-2026-0203","name":"Gemini 3.1 Flash-Lite","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"3 March 2026","url":"/safety/models/MV-2026-0203"},"domain":"other","domain_label":"Other","level":"CCL (Critical Capability Level)","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessments have shown that gemini 3.1 flash-lite is less capable than gemini 3.1 pro, therefore based on gemini 3.1 pro, we are confident that gemini 3.1 flash-lite is also unlikely to reach any ccls.","publisher":"Google DeepMind","date":"3 March 2026","date_iso":"2026-03-03","url":"https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/","text":"our assessments have shown that gemini 3.1 flash-lite is less capable than gemini 3.1 pro, therefore based on gemini 3.1 pro, we are confident that gemini 3.1 flash-lite is also unlikely to reach any ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0203#threshold-4bde6465d5","anchor":"threshold-4bde6465d5"}]},{"version":{"id":"MV-2026-0205","name":"Gemini 3.1 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"19 February 2026","url":"/safety/models/MV-2026-0205"},"rows":[{"version":{"id":"MV-2026-0205","name":"Gemini 3.1 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"19 February 2026","url":"/safety/models/MV-2026-0205"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Uplift level 1","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we found that the model remains below alert thresholds for the cbrn, harmful manipulation, machine learning r&d, and misalignment ccls.","publisher":"Google DeepMind","date":"19 February 2026","date_iso":"2026-02-19","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Pro-Model-Card.pdf","text":"we found that the model remains below alert thresholds for the cbrn, harmful manipulation, machine learning r&d, and misalignment ccls.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0205#threshold-3fd67b7dac","anchor":"threshold-3fd67b7dac"},{"version":{"id":"MV-2026-0205","name":"Gemini 3.1 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"19 February 2026","url":"/safety/models/MV-2026-0205"},"domain":"cyber","domain_label":"Cyber","level":"Uplift level 1","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"as previous models passed the alert threshold for cyber, we performed more additional testing in this domain on gemini 3.1 pro with and without deep think mode, and found that the model remains below the cyber ccl.","publisher":"Google DeepMind","date":"19 February 2026","date_iso":"2026-02-19","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Pro-Model-Card.pdf","text":"as previous models passed the alert threshold for cyber, we performed more additional testing in this domain on gemini 3.1 pro with and without deep think mode, and found that the model remains below the cyber ccl.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0205#threshold-9c774352dc","anchor":"threshold-9c774352dc"}]},{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"rows":[{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude sonnet 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","publisher":"Anthropic","date":"17 February 2026","date_iso":"2026-02-17","url":"https://www.anthropic.com/claude-sonnet-4-6-system-card","text":"our determination is that claude sonnet 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0208#threshold-6b840577c6","anchor":"threshold-6b840577c6"},{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"claude sonnet 4.6 performed above thresholds for asl-3 rule-in on all three evaluations for asl-3, and it did not exceed the performance of previous models.","publisher":"Anthropic","date":"17 February 2026","date_iso":"2026-02-17","url":"https://www.anthropic.com/claude-sonnet-4-6-system-card","text":"claude sonnet 4.6 performed above thresholds for asl-3 rule-in on all three evaluations for asl-3, and it did not exceed the performance of previous models.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0208#threshold-e6b4af94f2","anchor":"threshold-e6b4af94f2"},{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"in particular, it did not cross the threshold on our asl-4 rule-out evaluation short-horizon computational biology tasks evaluation.","publisher":"Anthropic","date":"17 February 2026","date_iso":"2026-02-17","url":"https://www.anthropic.com/claude-sonnet-4-6-system-card","text":"in particular, it did not cross the threshold on our asl-4 rule-out evaluation short-horizon computational biology tasks evaluation.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0208#threshold-aa23cfe1d1","anchor":"threshold-aa23cfe1d1"},{"version":{"id":"MV-2026-0208","name":"Claude Sonnet 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"17 February 2026","released_iso":"2026-02-17","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0208"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CBRN-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude sonnet 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","publisher":"Anthropic","date":"17 February 2026","date_iso":"2026-02-17","url":"https://www.anthropic.com/claude-sonnet-4-6-system-card","text":"our determination is that claude sonnet 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0208#threshold-6104bf9d7d","anchor":"threshold-6104bf9d7d"}]},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"rows":[{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude opus 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"our determination is that claude opus 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-0012ab64b5","anchor":"threshold-0012ab64b5"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-4","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"for ai r&d capabilities, we found that claude opus 4.6 has saturated most of our automated evaluations, meaning they no longer provide useful evidence for ruling out asl-4 level autonomy.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"for ai r&d capabilities, we found that claude opus 4.6 has saturated most of our automated evaluations, meaning they no longer provide useful evidence for ruling out asl-4 level autonomy.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-b06b179120","anchor":"threshold-b06b179120"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"the model crossed or met thresholds on all asl-3 evaluations except our synthesis screening evasion, consistent with incremental capability improvements driven primarily by better agentic workflows.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"the model crossed or met thresholds on all asl-3 evaluations except our synthesis screening evasion, consistent with incremental capability improvements driven primarily by better agentic workflows.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-e24763e7fa","anchor":"threshold-e24763e7fa"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"claude opus 4.6 is our strongest biology model to date and demonstrates meaningful capability improvements, but we do not believe it merits asl-4 safeguards.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"claude opus 4.6 is our strongest biology model to date and demonstrates meaningful capability improvements, but we do not believe it merits asl-4 safeguards.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-556fc6a3bb","anchor":"threshold-556fc6a3bb"},{"version":{"id":"MV-2026-0217","name":"Claude Opus 4.6","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"5 February 2026","url":"/safety/models/MV-2026-0217"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CBRN-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude opus 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","publisher":"Anthropic","date":"5 February 2026","date_iso":"2026-02-05","url":"https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf","text":"our determination is that claude opus 4.6 does not cross either the ai r&d-4 or the cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0217#threshold-32187b620a","anchor":"threshold-32187b620a"}]},{"version":{"id":"MV-2026-0216","name":"GPT-5.3-Codex","provider":"OpenAI","provider_slug":"openai","released":"5 February 2026","released_iso":"2026-02-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0216"},"rows":[{"version":{"id":"MV-2026-0216","name":"GPT-5.3-Codex","provider":"OpenAI","provider_slug":"openai","released":"5 February 2026","released_iso":"2026-02-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0216"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"it does not reach high capability on ai self-improvement.","publisher":"OpenAI","date":"5 February 2026","date_iso":"2026-02-05","url":"https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf","text":"it does not reach high capability on ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0216#threshold-dfa617b4dc","anchor":"threshold-dfa617b4dc"},{"version":{"id":"MV-2026-0216","name":"GPT-5.3-Codex","provider":"OpenAI","provider_slug":"openai","released":"5 February 2026","released_iso":"2026-02-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0216"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"we are also treating gpt-5.3-codex as high risk in the biological and chemical domain, and applying the same safeguards as for our other deployments of models that reach this designation.","publisher":"OpenAI","date":"5 February 2026","date_iso":"2026-02-05","url":"https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf","text":"we are also treating gpt-5.3-codex as high risk in the biological and chemical domain, and applying the same safeguards as for our other deployments of models that reach this designation.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0216#threshold-5bccffb03a","anchor":"threshold-5bccffb03a"},{"version":{"id":"MV-2026-0216","name":"GPT-5.3-Codex","provider":"OpenAI","provider_slug":"openai","released":"5 February 2026","released_iso":"2026-02-05","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2026-0216"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"this is the first launch we are treating as high capability in the cybersecurity domain under our preparedness framework, activating the associated safeguards.","publisher":"OpenAI","date":"5 February 2026","date_iso":"2026-02-05","url":"https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf","text":"this is the first launch we are treating as high capability in the cybersecurity domain under our preparedness framework, activating the associated safeguards.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2026-0216#threshold-3022fdc78e","anchor":"threshold-3022fdc78e"}]},{"version":{"id":"MV-2025-0005","name":"Claude Opus 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"November 2025","url":"/safety/models/MV-2025-0005"},"rows":[{"version":{"id":"MV-2025-0005","name":"Claude Opus 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"November 2025","url":"/safety/models/MV-2025-0005"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our determination is that claude opus 4.5 does not cross either the ai r&d-4 or cbrn-4 capability threshold.","publisher":"Anthropic","date":"November 2025","date_iso":"2025-11","url":"https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf","text":"our determination is that claude opus 4.5 does not cross either the ai r&d-4 or cbrn-4 capability threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0005#threshold-8bc97337aa","anchor":"threshold-8bc97337aa"},{"version":{"id":"MV-2025-0005","name":"Claude Opus 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"November 2025","url":"/safety/models/MV-2025-0005"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"CBRN-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we determine that claude opus 4.5 does not cross the cbrn-4 threshold.","publisher":"Anthropic","date":"November 2025","date_iso":"2025-11","url":"https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf","text":"we determine that claude opus 4.5 does not cross the cbrn-4 threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0005#threshold-c188f9c9cf","anchor":"threshold-c188f9c9cf"}]},{"version":{"id":"MV-2025-0004","name":"Gemini 3 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"November 2025","released_iso":"2025-11","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0004"},"rows":[{"version":{"id":"MV-2025-0004","name":"Gemini 3 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"November 2025","released_iso":"2025-11","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0004"},"domain":"cyber","domain_label":"Cyber","level":"Uplift level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"cybersecurity on key skills benchmark, v1 hard challenges: 11/12 challenges solved; v2 challenges: 0/13 solved end-to-end.","publisher":"Google DeepMind","date":"November 2025","date_iso":"2025-11","url":"https://deepmind.google/models/model-cards/gemini-3-pro/","text":"cybersecurity on key skills benchmark, v1 hard challenges: 11/12 challenges solved; v2 challenges: 0/13 solved end-to-end.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0004#threshold-80aeac0c5c","anchor":"threshold-80aeac0c5c"},{"version":{"id":"MV-2025-0004","name":"Gemini 3 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"November 2025","released_iso":"2025-11","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0004"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Instrumental reasoning level 1 + 2","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"misalignment (exploratory) agent solves 3/11 situational awareness challenges and 1/4 stealth challenges.","publisher":"Google DeepMind","date":"November 2025","date_iso":"2025-11","url":"https://deepmind.google/models/model-cards/gemini-3-pro/","text":"misalignment (exploratory) agent solves 3/11 situational awareness challenges and 1/4 stealth challenges.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0004#threshold-03acb1b1a4","anchor":"threshold-03acb1b1a4"},{"version":{"id":"MV-2025-0004","name":"Gemini 3 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"November 2025","released_iso":"2025-11","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0004"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Level 1 (exploratory)","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"harmful manipulation model manipulative efficacy improves on non-generative ai baseline, but shows no significant uplift versus prior models and does not reach alert thresholds.","publisher":"Google DeepMind","date":"November 2025","date_iso":"2025-11","url":"https://deepmind.google/models/model-cards/gemini-3-pro/","text":"harmful manipulation model manipulative efficacy improves on non-generative ai baseline, but shows no significant uplift versus prior models and does not reach alert thresholds.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0004#threshold-a5082992ed","anchor":"threshold-a5082992ed"}]},{"version":{"id":"MV-2025-0007","name":"Claude Haiku 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"October 2025","url":"/safety/models/MV-2025-0007"},"rows":[{"version":{"id":"MV-2025-0007","name":"Claude Haiku 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"October 2025","url":"/safety/models/MV-2025-0007"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-3","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our evaluation results showed that claude haiku 4.5 remained well below asl-3 thresholds across all domains of concern.","publisher":"Anthropic","date":"October 2025","date_iso":"2025-10","url":"https://www-cdn.anthropic.com/7aad69bf12627d42234e01ee7c36305dc2f6a970.pdf","text":"our evaluation results showed that claude haiku 4.5 remained well below asl-3 thresholds across all domains of concern.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0007#threshold-388ca4765f","anchor":"threshold-388ca4765f"},{"version":{"id":"MV-2025-0007","name":"Claude Haiku 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"October 2025","url":"/safety/models/MV-2025-0007"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"our evaluation results showed that claude haiku 4.5 remained well below asl-3 thresholds across all domains of concern.","publisher":"Anthropic","date":"October 2025","date_iso":"2025-10","url":"https://www-cdn.anthropic.com/7aad69bf12627d42234e01ee7c36305dc2f6a970.pdf","text":"our evaluation results showed that claude haiku 4.5 remained well below asl-3 thresholds across all domains of concern.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0007#threshold-04f4ae624a","anchor":"threshold-04f4ae624a"}]},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"rows":[{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"AI R&D-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"all models tested remain below our pre-committed rule-out threshold of 0.6.","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"all models tested remain below our pre-committed rule-out threshold of 0.6.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-b48be0f714","anchor":"threshold-b48be0f714"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"exhibited improved autonomy capabilities in software engineering and ai research tasks, though still below the asl-4 rule-out threshold.","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"exhibited improved autonomy capabilities in software engineering and ai research tasks, though still below the asl-4 rule-out threshold.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-66a676ff96","anchor":"threshold-66a676ff96"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"remained well below asl-4 thresholds across all domains of concern;","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"remained well below asl-4 thresholds across all domains of concern;","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-21b885ffab","anchor":"threshold-21b885ffab"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4 rule-out","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"however, given that this type of logic can have a general frog-boiling effect—where gradual changes mask the crossing of important thresholds—we do not use this evaluation as indicative of a strong asl-4 rule-out.","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"however, given that this type of logic can have a general frog-boiling effect—where gradual changes mask the crossing of important thresholds—we do not use this evaluation as indicative of a strong asl-4 rule-out.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-c68b5f8874","anchor":"threshold-c68b5f8874"},{"version":{"id":"MV-2025-0012","name":"Claude Sonnet 4.5","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"September 2025","url":"/safety/models/MV-2025-0012"},"domain":"cyber","domain_label":"Cyber","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"showed meaningful improvements in cyber capabilities, particularly in vulnerability discovery and code analysis;","publisher":"Anthropic","date":"September 2025","date_iso":"2025-09","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf","text":"showed meaningful improvements in cyber capabilities, particularly in vulnerability discovery and code analysis;","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0012#threshold-893d0f3495","anchor":"threshold-893d0f3495"}]},{"version":{"id":"MV-2025-0014","name":"GPT-5","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"13 August 2025","provider_doc_dated":"","url":"/safety/models/MV-2025-0014"},"rows":[{"version":{"id":"MV-2025-0014","name":"GPT-5","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"13 August 2025","provider_doc_dated":"","url":"/safety/models/MV-2025-0014"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"crossed","outcome_label":"Stated as crossed","cite":{"quote":"similarly to chatgpt agent, we have decided to treat gpt-5-thinking as high capability in the biological and chemical domain under our preparedness framework, activating the associated safeguards.","publisher":"OpenAI","date":"13 August 2025","date_iso":"2025-08-13","url":"https://cdn.openai.com/gpt-5-system-card.pdf","text":"similarly to chatgpt agent, we have decided to treat gpt-5-thinking as high capability in the biological and chemical domain under our preparedness framework, activating the associated safeguards.","damaged":false,"lowered":true},"also":[{"quote":"we decided to treat this launch as high capability in the biological and chemical domain, activating the associated preparedness safeguards.","publisher":"OpenAI","date":"13 August 2025","date_iso":"2025-08-13","url":"https://cdn.openai.com/gpt-5-system-card.pdf","text":"we decided to treat this launch as high capability in the biological and chemical domain, activating the associated preparedness safeguards.","damaged":false,"lowered":true}],"anchor_url":"/safety/models/MV-2025-0014#threshold-f2c6b00a38","anchor":"threshold-f2c6b00a38"},{"version":{"id":"MV-2025-0014","name":"GPT-5","provider":"OpenAI","provider_slug":"openai","released":"","released_iso":"","card_dated":"13 August 2025","provider_doc_dated":"","url":"/safety/models/MV-2025-0014"},"domain":"cyber","domain_label":"Cyber","level":"High cyber risk","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"the gpt-5 model series does not meet the threshold for high cyber risk.","publisher":"OpenAI","date":"13 August 2025","date_iso":"2025-08-13","url":"https://cdn.openai.com/gpt-5-system-card.pdf","text":"the gpt-5 model series does not meet the threshold for high cyber risk.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0014#threshold-59ada166b8","anchor":"threshold-59ada166b8"}]},{"version":{"id":"MV-2025-0017","name":"Claude Opus 4.1","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0017"},"rows":[{"version":{"id":"MV-2025-0017","name":"Claude Opus 4.1","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0017"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","publisher":"Anthropic","date":"August 2025","date_iso":"2025-08","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf","text":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0017#threshold-eb476b7fde","anchor":"threshold-eb476b7fde"},{"version":{"id":"MV-2025-0017","name":"Claude Opus 4.1","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0017"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","publisher":"Anthropic","date":"August 2025","date_iso":"2025-08","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf","text":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0017#threshold-0d0967fc34","anchor":"threshold-0d0967fc34"},{"version":{"id":"MV-2025-0017","name":"Claude Opus 4.1","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0017"},"domain":"cyber","domain_label":"Cyber","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","publisher":"Anthropic","date":"August 2025","date_iso":"2025-08","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf","text":"since claude opus 4.1 is deployed with asl-3 protections, we focused on confirming it remains well below asl-4 thresholds across cbrn (chemical, biological, radiological, and nuclear), cyber, and autonomy domains.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0017#threshold-62c30509fe","anchor":"threshold-62c30509fe"}]},{"version":{"id":"MV-2025-0016","name":"Gemini 2.5 Deep Think","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0016"},"rows":[{"version":{"id":"MV-2025-0016","name":"Gemini 2.5 Deep Think","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0016"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Uplift level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"our assessment is that the model has enough technical knowledge in certain cbrn scenarios and stages to be considered at early warning.","publisher":"Google DeepMind","date":"August 2025","date_iso":"2025-08","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf","text":"our assessment is that the model has enough technical knowledge in certain cbrn scenarios and stages to be considered at early warning.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0016#threshold-ad43754209","anchor":"threshold-ad43754209"},{"version":{"id":"MV-2025-0016","name":"Gemini 2.5 Deep Think","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0016"},"domain":"cyber","domain_label":"Cyber","level":"Cyber Uplift Level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"ccl reached? no, but the early warning alert threshold was originally reached by gemini 2.5 pro and continues to be met by gemini 2.5 deep think.","publisher":"Google DeepMind","date":"August 2025","date_iso":"2025-08","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf","text":"ccl reached? no, but the early warning alert threshold was originally reached by gemini 2.5 pro and continues to be met by gemini 2.5 deep think.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0016#threshold-9eb73a502c","anchor":"threshold-9eb73a502c"},{"version":{"id":"MV-2025-0016","name":"Gemini 2.5 Deep Think","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"August 2025","url":"/safety/models/MV-2025-0016"},"domain":"cyber","domain_label":"Cyber","level":"Uplift level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"for cyber uplift level 1 , the early warning alert threshold was originally reached by gemini 2.5 pro and continues to be met by gemini 2.5 deep think.","publisher":"Google DeepMind","date":"August 2025","date_iso":"2025-08","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf","text":"for cyber uplift level 1 , the early warning alert threshold was originally reached by gemini 2.5 pro and continues to be met by gemini 2.5 deep think.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0016#threshold-6e73c3fdac","anchor":"threshold-6e73c3fdac"}]},{"version":{"id":"MV-2025-0019","name":"Gemini 2.5 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"27 June 2025","url":"/safety/models/MV-2025-0019"},"rows":[{"version":{"id":"MV-2025-0019","name":"Gemini 2.5 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"27 June 2025","url":"/safety/models/MV-2025-0019"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Machine Learning R&D Uplift Level 1","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"given gemini 2.5 pro preview's poor median performance on re-bench relative to experts, the model likely lacks the necessary capabilities to automate or significantly uplift any significant fraction of the research process.","publisher":"Google DeepMind","date":"27 June 2025","date_iso":"2025-06-27","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf","text":"given gemini 2.5 pro preview's poor median performance on re-bench relative to experts, the model likely lacks the necessary capabilities to automate or significantly uplift any significant fraction of the research process.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0019#threshold-e0e02cc323","anchor":"threshold-e0e02cc323"},{"version":{"id":"MV-2025-0019","name":"Gemini 2.5 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"27 June 2025","url":"/safety/models/MV-2025-0019"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Instrumental Reasoning Level 1","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"instru- mental reasoning level 1 ✘ ccl not reached","publisher":"Google DeepMind","date":"27 June 2025","date_iso":"2025-06-27","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf","text":"instru- mental reasoning level 1 ✘ ccl not reached","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0019#threshold-02e22335e0","anchor":"threshold-02e22335e0"},{"version":{"id":"MV-2025-0019","name":"Gemini 2.5 Pro","provider":"Google DeepMind","provider_slug":"google-deepmind","released":"","released_iso":"","card_dated":"","provider_doc_dated":"27 June 2025","url":"/safety/models/MV-2025-0019"},"domain":"deception_and_sabotage","domain_label":"Deception & sabotage","level":"Instrumental Reasoning Level 2","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"instru- mental reasoning level 2 ✘ ccl not reached","publisher":"Google DeepMind","date":"27 June 2025","date_iso":"2025-06-27","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf","text":"instru- mental reasoning level 2 ✘ ccl not reached","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0019#threshold-751da4b11d","anchor":"threshold-751da4b11d"}]},{"version":{"id":"MV-2025-0023","name":"Claude Opus 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0023"},"rows":[{"version":{"id":"MV-2025-0023","name":"Claude Opus 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0023"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"for claude opus 4, our evaluations showed that, whereas we could confidently rule out asl-4 capabilities, we could not conclusively rule out asl-3 capabilities—which led to our precautionary decision to deploy it with asl-3 protections .","publisher":"Anthropic","date":"4 May 2025","date_iso":"2025-05-04","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf","text":"for claude opus 4, our evaluations showed that, whereas we could confidently rule out asl-4 capabilities, we could not conclusively rule out asl-3 capabilities—which led to our precautionary decision to deploy it with asl-3 protections .","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0023#threshold-0c91007de2","anchor":"threshold-0c91007de2"},{"version":{"id":"MV-2025-0023","name":"Claude Opus 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0023"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"for claude opus 4, our evaluations showed that, whereas we could confidently rule out asl-4 capabilities, we could not conclusively rule out asl-3 capabilities—which led to our precautionary decision to deploy it with asl-3 protections .","publisher":"Anthropic","date":"4 May 2025","date_iso":"2025-05-04","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf","text":"for claude opus 4, our evaluations showed that, whereas we could confidently rule out asl-4 capabilities, we could not conclusively rule out asl-3 capabilities—which led to our precautionary decision to deploy it with asl-3 protections .","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0023#threshold-99e3b55ca9","anchor":"threshold-99e3b55ca9"}]},{"version":{"id":"MV-2025-0022","name":"Claude Sonnet 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0022"},"rows":[{"version":{"id":"MV-2025-0022","name":"Claude Sonnet 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0022"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-3","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","publisher":"Anthropic","date":"4 May 2025","date_iso":"2025-05-04","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf","text":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0022#threshold-52087ecd28","anchor":"threshold-52087ecd28"},{"version":{"id":"MV-2025-0022","name":"Claude Sonnet 4","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"4 May 2025","url":"/safety/models/MV-2025-0022"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"ASL-4","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","publisher":"Anthropic","date":"4 May 2025","date_iso":"2025-05-04","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf","text":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0022#threshold-14e15bb7c5","anchor":"threshold-14e15bb7c5"}]},{"version":{"id":"MV-2025-0024","name":"o3","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0024"},"rows":[{"version":{"id":"MV-2025-0024","name":"o3","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0024"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0024#threshold-b96c76e329","anchor":"threshold-b96c76e329"},{"version":{"id":"MV-2025-0024","name":"o3","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0024"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0024#threshold-465fa69dad","anchor":"threshold-465fa69dad"},{"version":{"id":"MV-2025-0024","name":"o3","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0024"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0024#threshold-ac9679fd1d","anchor":"threshold-ac9679fd1d"}]},{"version":{"id":"MV-2025-0025","name":"o4-mini","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0025"},"rows":[{"version":{"id":"MV-2025-0025","name":"o4-mini","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0025"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0025#threshold-6a2fa201dd","anchor":"threshold-6a2fa201dd"},{"version":{"id":"MV-2025-0025","name":"o4-mini","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0025"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0025#threshold-7015fbd21a","anchor":"threshold-7015fbd21a"},{"version":{"id":"MV-2025-0025","name":"o4-mini","provider":"OpenAI","provider_slug":"openai","released":"16 April 2025","released_iso":"2025-04-16","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0025"},"domain":"cyber","domain_label":"Cyber","level":"High","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","publisher":"OpenAI","date":"16 April 2025","date_iso":"2025-04-16","url":"https://openai.com/index/o3-o4-mini-system-card/","text":"openai’s safety advisory group (sag) reviewed the results of our preparedness evaluations and determined that openai o3 and o4-mini do not reach the high threshold in any of our three tracked categories: biological and chemical capability, cybersecurity, and ai self-improvement.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0025#threshold-414d403dca","anchor":"threshold-414d403dca"}]},{"version":{"id":"MV-2025-0039","name":"Claude Sonnet 3.7","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"February 2025","url":"/safety/models/MV-2025-0039"},"rows":[{"version":{"id":"MV-2025-0039","name":"Claude Sonnet 3.7","provider":"Anthropic","provider_slug":"anthropic","released":"","released_iso":"","card_dated":"","provider_doc_dated":"February 2025","url":"/safety/models/MV-2025-0039"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"ASL-4","outcome":"undetermined","outcome_label":"Not determined","cite":{"quote":"these problems are harder to solve than the previous problems and we consider them as part of our asl-4 evaluations.","publisher":"Anthropic","date":"February 2025","date_iso":"2025-02","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf","text":"these problems are harder to solve than the previous problems and we consider them as part of our asl-4 evaluations.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0039#threshold-de8350a522","anchor":"threshold-de8350a522"}]},{"version":{"id":"MV-2025-0040","name":"Operator","provider":"OpenAI","provider_slug":"openai","released":"23 January 2025","released_iso":"2025-01-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0040"},"rows":[{"version":{"id":"MV-2025-0040","name":"Operator","provider":"OpenAI","provider_slug":"openai","released":"23 January 2025","released_iso":"2025-01-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0040"},"domain":"autonomy_and_replication","domain_label":"Autonomy & replication","level":"Low","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"we adapted these two evals for the computer-use setting and found the pre-mitigation operator model to be “low” risk in both categories, consistent with the gpt‑4o model.","publisher":"OpenAI","date":"23 January 2025","date_iso":"2025-01-23","url":"https://openai.com/index/operator-system-card/","text":"we adapted these two evals for the computer-use setting and found the pre-mitigation operator model to be “low” risk in both categories, consistent with the gpt‐4o model.","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0040#threshold-7cd9943023","anchor":"threshold-7cd9943023"},{"version":{"id":"MV-2025-0040","name":"Operator","provider":"OpenAI","provider_slug":"openai","released":"23 January 2025","released_iso":"2025-01-23","card_dated":"","provider_doc_dated":"","url":"/safety/models/MV-2025-0040"},"domain":"cbrn","domain_label":"Chemical, biological, radiological and nuclear","level":"Low","outcome":"not_crossed","outcome_label":"Stated as not crossed","cite":{"quote":"this 1% qualifies as “low” risk, matching gpt‑4o’s “low” risk level, as detailed in openai’s preparedness framework (opens in a new window) 3 .","publisher":"OpenAI","date":"23 January 2025","date_iso":"2025-01-23","url":"https://openai.com/index/operator-system-card/","text":"this 1% qualifies as “low” risk, matching gpt‐4o’s “low” risk level, as detailed in openai’s preparedness framework (opens in a new window) 3 .","damaged":false,"lowered":true},"also":[],"anchor_url":"/safety/models/MV-2025-0040#threshold-f70f98f3fb","anchor":"threshold-f70f98f3fb"}]}],"terms":[{"term":"CCL","words":"Critical Capability Level","plain":"A capability level at which, without safeguards, a model may pose a heightened risk of severe harm.","quote":"capability thresholds called “critical capability levels (ccls).” these are capability levels at which, absent mitigation measures, frontier ai models or systems may pose heightened risk of severe harm.","publisher":"Google DeepMind","url":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf"},{"term":"TCL","words":"Tracked Capability Level","plain":"A capability level for significant risks that may appear at a lower capability than a Critical Capability Level.","quote":"in v3.1 of the framework, we introduced “tracked capability levels (tcls).” tcls are meant to capture significant risks that may manifest at a lower capability threshold than our ccls.","publisher":"Google DeepMind","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf"},{"term":"ASL","words":"AI Safety Level","plain":"A graded set of safety and security measures that become stricter as model capabilities increase.","quote":"we use AI Safety Level Standards (ASL Standards), graduated sets of safety and security measures that become more stringent as model capabilities increase.","publisher":"Anthropic","url":"https://www.anthropic.com/news/announcing-our-updated-responsible-scaling-policy"},{"term":"FAIF","words":"Frontier AI Framework","plain":"xAI's framework of risk domains, capability thresholds and the mitigations that go with them.","quote":"this frontier ai framework (“ faif ”) outlines xai’s approach to policies for mitigation of significant risks associated with the development, deployment, and release of xai’s frontier ai models, such as grok.","publisher":"xAI","url":"https://media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-99c40684.pdf"}],"states":[{"key":"crossed","label":"Stated as crossed","count":20,"url":"/safety/crossings?state=crossed","active":false},{"key":"not_crossed","label":"Stated as not crossed","count":80,"url":"/safety/crossings?state=not_crossed","active":false},{"key":"undetermined","label":"Not determined","count":43,"url":"/safety/crossings?state=undetermined","active":false}],"domains":[{"key":"autonomy_and_replication","label":"Autonomy & replication","count":41,"url":"/safety/crossings?domain=autonomy_and_replication&state=all","active":false},{"key":"cbrn","label":"Chemical, biological, radiological and nuclear","count":57,"url":"/safety/crossings?domain=cbrn&state=all","active":false},{"key":"cyber","label":"Cyber","count":28,"url":"/safety/crossings?domain=cyber&state=all","active":false},{"key":"deception_and_sabotage","label":"Deception & sabotage","count":7,"url":"/safety/crossings?domain=deception_and_sabotage&state=all","active":false},{"key":"misuse_and_jailbreak","label":"Misuse & jailbreak","count":2,"url":"/safety/crossings?domain=misuse_and_jailbreak&state=all","active":false},{"key":"other","label":"Other","count":8,"url":"/safety/crossings?domain=other&state=all","active":false}],"clear_url":"/safety/crossings?state=all","crossed_url":"/safety/crossings?state=crossed","all_url":"/safety/crossings?state=all","all_count":143,"group_first":3}