{"total":258,"limit":100,"offset":0,"determinations":[{"version_id":"MV-2025-0004","provider":"Google DeepMind","domain":"cbrn","framework_id":null,"level":"Uplift level 1","rank":null,"determination":"undetermined","quote":"CBRN Gemini 3 Pro provides accurate and occasionally actionable information but generally fails to offer novel or sufficiently complete and detailed instructions to significantly enhance the capabilities of low to medium resourced threat actors.","url":"https://deepmind.google/models/model-cards/gemini-3-pro/"},{"version_id":"MV-2025-0004","provider":"Google DeepMind","domain":"cyber","framework_id":null,"level":"Uplift level 1","rank":null,"determination":"undetermined","quote":"Cybersecurity On key skills benchmark, v1 hard challenges: 11/12 challenges solved; v2 challenges: 0/13 solved end-to-end.","url":"https://deepmind.google/models/model-cards/gemini-3-pro/"},{"version_id":"MV-2025-0004","provider":"Google DeepMind","domain":"deception_and_sabotage","framework_id":null,"level":"Level 1 (exploratory)","rank":null,"determination":"not_crossed","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.","url":"https://deepmind.google/models/model-cards/gemini-3-pro/"},{"version_id":"MV-2025-0004","provider":"Google DeepMind","domain":"autonomy_and_replication","framework_id":null,"level":"Acceleration level 1","rank":null,"determination":"undetermined","quote":"Machine Learning R&D Gemini 3 Pro performs better than Gemini 2.5 models, especially on the Scaling Law Experiment and Optimize LLM Foundry tasks in RE-Bench (Wijk et al., 2024).","url":"https://deepmind.google/models/model-cards/gemini-3-pro/"},{"version_id":"MV-2025-0004","provider":"Google DeepMind","domain":"autonomy_and_replication","framework_id":null,"level":"Automation level 1","rank":null,"determination":"undetermined","quote":"Machine Learning R&D Gemini 3 Pro performs better than Gemini 2.5 models, especially on the Scaling Law Experiment and Optimize LLM Foundry tasks in RE-Bench (Wijk et al., 2024).","url":"https://deepmind.google/models/model-cards/gemini-3-pro/"},{"version_id":"MV-2025-0004","provider":"Google DeepMind","domain":"deception_and_sabotage","framework_id":null,"level":"Instrumental reasoning levels 1 + 2","rank":null,"determination":"undetermined","quote":"Misalignment (Exploratory) Agent solves 3/11 situational awareness challenges and 1/4 stealth challenges.","url":"https://deepmind.google/models/model-cards/gemini-3-pro/"},{"version_id":"MV-2025-0005","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":"FW-0004","level":"AI R&D-4","rank":1,"determination":"not_crossed","quote":"Our determination is that Claude Opus 4.5 does not cross either the AI R&D-4 or CBRN-4 capability threshold.","url":"https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf"},{"version_id":"MV-2025-0005","provider":"Anthropic","domain":"cbrn","framework_id":"FW-0004","level":"CBRN-4","rank":2,"determination":"not_crossed","quote":"We determine that Claude Opus 4.5 does not cross the CBRN-4 threshold.","url":"https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf"},{"version_id":"MV-2025-0005","provider":"Anthropic","domain":"cbrn","framework_id":"FW-0015","level":"ASL-3","rank":1,"determination":"undetermined","quote":"RSP safeguards applied to Claude Opus 4.5: AI Safety Level 3 (ASL-3)","url":"https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf"},{"version_id":"MV-2025-0005","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-4","rank":null,"determination":"not_crossed","quote":"As a result, whereas 119 Claude Opus 4.5 is our strongest biology model to date, we do not think it merits ASL-4 safeguards.","url":"https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf"},{"version_id":"MV-2025-0007","provider":"Anthropic","domain":"cbrn","framework_id":"FW-0015","level":"ASL-3","rank":1,"determination":"not_crossed","quote":"Our evaluation results showed that Claude Haiku 4.5 remained well below ASL-3 thresholds across all domains of concern.","url":"https://www-cdn.anthropic.com/7aad69bf12627d42234e01ee7c36305dc2f6a970.pdf"},{"version_id":"MV-2025-0007","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"ASL-3","rank":null,"determination":"not_crossed","quote":"Our evaluation results showed that Claude Haiku 4.5 remained well below ASL-3 thresholds across all domains of concern.","url":"https://www-cdn.anthropic.com/7aad69bf12627d42234e01ee7c36305dc2f6a970.pdf"},{"version_id":"MV-2025-0012","provider":"Anthropic","domain":"cbrn","framework_id":"FW-0015","level":"ASL-3","rank":1,"determination":"undetermined","quote":"Remained well below ASL-4 thresholds across all domains of concern;","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf"},{"version_id":"MV-2025-0012","provider":"Anthropic","domain":"cyber","framework_id":null,"level":"ASL-3","rank":null,"determination":"undetermined","quote":"Showed meaningful improvements in cyber capabilities, particularly in vulnerability discovery and code analysis;","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf"},{"version_id":"MV-2025-0012","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"ASL-3","rank":null,"determination":"undetermined","quote":"Exhibited improved autonomy capabilities in software engineering and AI research tasks, though still below the ASL-4 rule-out threshold.","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf"},{"version_id":"MV-2025-0012","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-4","rank":null,"determination":"not_crossed","quote":"While Claude Sonnet 4.5 demonstrated improved performance with respect to Claude Opus 4 and Claude Opus 4.1, it nevertheless remained below our ASL-4 rule-out thresholds.","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf"},{"version_id":"MV-2025-0012","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-4 rule-out","rank":null,"determination":"undetermined","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.","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf"},{"version_id":"MV-2025-0012","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"ASL-4","rank":null,"determination":"not_crossed","quote":"the combination of performance below threshold in AI R&D Suite 2 and the survey results indicated that Claude Sonnet 4.5’s capabilities remained 138 below the ASL-4 threshold of fully automating the work of an entry-level, remote-only researcher at Anthropic.","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf"},{"version_id":"MV-2025-0012","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":"FW-0004","level":"AI R&D-4","rank":1,"determination":"not_crossed","quote":"All models tested remain below our pre-committed rule-out threshold of 0.6.","url":"https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf"},{"version_id":"MV-2025-0014","provider":"OpenAI","domain":"cbrn","framework_id":null,"level":"high capability","rank":null,"determination":"crossed","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.","url":"https://cdn.openai.com/gpt-5-system-card.pdf"},{"version_id":"MV-2025-0014","provider":"OpenAI","domain":"cbrn","framework_id":null,"level":"high","rank":null,"determination":"crossed","quote":"We decided to treat this launch as High capability in the Biological and Chemical domain, activating the associated Preparedness safeguards.","url":"https://cdn.openai.com/gpt-5-system-card.pdf"},{"version_id":"MV-2025-0014","provider":"OpenAI","domain":"cyber","framework_id":null,"level":"high cyber risk","rank":null,"determination":"not_crossed","quote":"The gpt-5 model series does not meet the threshold for high cyber risk.","url":"https://cdn.openai.com/gpt-5-system-card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"cbrn","framework_id":null,"level":"uplift level 1","rank":null,"determination":"undetermined","quote":"Our assessment is that the model has enough technical knowledge in certain CBRN scenarios and stages to be considered at early warning.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"cyber","framework_id":null,"level":"uplift level 1","rank":null,"determination":"undetermined","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.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"autonomy_and_replication","framework_id":null,"level":"autonomy level 1","rank":null,"determination":"not_crossed","quote":"Autonomy Level 1 CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"autonomy_and_replication","framework_id":null,"level":"uplift level 1","rank":null,"determination":"not_crossed","quote":"Uplift Level 1 CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"autonomy_and_replication","framework_id":null,"level":"machine learning r&d uplift level 1","rank":null,"determination":"not_crossed","quote":"Machine Learning R&D Uplift Level 1 CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"deception_and_sabotage","framework_id":"FW-0002","level":"instrumental reasoning level 1","rank":1,"determination":"not_crossed","quote":"Instrumental Reasoning Level 1 CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"deception_and_sabotage","framework_id":"FW-0002","level":"instrumental reasoning level 2","rank":2,"determination":"not_crossed","quote":"Instrumental Reasoning Level 2 CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"cyber","framework_id":"FW-0002","level":"Cyber Uplift Level 1","rank":1,"determination":"undetermined","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.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0016","provider":"Google DeepMind","domain":"cyber","framework_id":null,"level":"cyber uplift level 1 ccl","rank":null,"determination":"undetermined","quote":"The model continues to perform at the early warning threshold level for the Cyber Uplift Level 1 CCL, and has reached the early warning alert threshold for CBRN Uplift Level 1 CCL.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Deep-Think-Model-Card.pdf"},{"version_id":"MV-2025-0017","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-4","rank":null,"determination":"not_crossed","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.","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf"},{"version_id":"MV-2025-0017","provider":"Anthropic","domain":"cyber","framework_id":null,"level":"ASL-4","rank":null,"determination":"not_crossed","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.","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf"},{"version_id":"MV-2025-0017","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"ASL-4","rank":null,"determination":"not_crossed","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.","url":"https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf"},{"version_id":"MV-2025-0019","provider":"Google DeepMind","domain":"cbrn","framework_id":null,"level":"uplift level 1","rank":null,"determination":"not_crossed","quote":"CBRN Uplift Level 1 CCL Definition : \"The model can be used to significantly assist a low-resourced actor with dual-use scientific protocols, resulting in a substantial increase in ability to cause a mass casualty event.\" 3 CCL reached? No.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf"},{"version_id":"MV-2025-0019","provider":"Google DeepMind","domain":"cyber","framework_id":null,"level":"uplift level 1","rank":null,"determination":"not_crossed","quote":"Uplift Level 1 ✘ CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf"},{"version_id":"MV-2025-0019","provider":"Google DeepMind","domain":"autonomy_and_replication","framework_id":null,"level":"autonomy level 1","rank":null,"determination":"not_crossed","quote":"Autonomy Level 1 ✘ CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf"},{"version_id":"MV-2025-0019","provider":"Google DeepMind","domain":"deception_and_sabotage","framework_id":"FW-0002","level":"instrumental reasoning level 1","rank":1,"determination":"not_crossed","quote":"Instru- mental Reasoning Level 1 ✘ CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf"},{"version_id":"MV-2025-0019","provider":"Google DeepMind","domain":"deception_and_sabotage","framework_id":"FW-0002","level":"instrumental reasoning level 2","rank":2,"determination":"not_crossed","quote":"Instru- mental Reasoning Level 2 ✘ CCL not reached","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf"},{"version_id":"MV-2025-0019","provider":"Google DeepMind","domain":"cyber","framework_id":"FW-0002","level":"cyber uplift level 1","rank":1,"determination":"not_crossed","quote":"We find that the model does not reach the Cyber Uplift Level 1 CCL.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf"},{"version_id":"MV-2025-0019","provider":"Google DeepMind","domain":"autonomy_and_replication","framework_id":null,"level":"machine learning r&d uplift level 1","rank":null,"determination":"undetermined","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.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf"},{"version_id":"MV-2025-0022","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"AI Safety Level 2","rank":null,"determination":"undetermined","quote":"informed by the testing described here, we have decided to deploy claude opus 4 under the ai safety level 3 standard and claude sonnet 4 under the ai safety level 2 standard.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0022","provider":"Anthropic","domain":"cyber","framework_id":null,"level":"AI Safety Level 2","rank":null,"determination":"undetermined","quote":"informed by the testing described here, we have decided to deploy claude opus 4 under the ai safety level 3 standard and claude sonnet 4 under the ai safety level 2 standard.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0022","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"AI Safety Level 2","rank":null,"determination":"undetermined","quote":"informed by the testing described here, we have decided to deploy claude opus 4 under the ai safety level 3 standard and claude sonnet 4 under the ai safety level 2 standard.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0022","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-2","rank":null,"determination":"undetermined","quote":"in contrast, we determined that claude sonnet 4 has not improved enough in the relevant capabilities to require the asl-3 standard.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0022","provider":"Anthropic","domain":"cbrn","framework_id":"FW-0015","level":"ASL-3","rank":1,"determination":"undetermined","quote":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0022","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-4","rank":null,"determination":"undetermined","quote":"specifically, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0022","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-3 rule-in threshold","rank":null,"determination":"undetermined","quote":"claude opus 4’s higher performance on both tasks increases our level of concern, while we believe that claude sonnet 4 remains below our rule-in threshold.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0022","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"ASL-4 autonomy","rank":null,"determination":"not_crossed","quote":"overall, whereas claude sonnet 4 shows capability improvements in select areas, its performance remains well below the thresholds that would warrant asl-4 protections.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0023","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ai safety level 3","rank":null,"determination":"undetermined","quote":"informed by the testing described here, we have decided to deploy claude opus 4 under the ai safety level 3 standard and claude sonnet 4 under the ai safety level 2 standard.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0023","provider":"Anthropic","domain":"cbrn","framework_id":"FW-0015","level":"ASL-3","rank":1,"determination":"undetermined","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 .","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0023","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-4","rank":null,"determination":"not_crossed","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 .","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0023","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-3 rule-in","rank":null,"determination":"undetermined","quote":"claude opus 4’s higher performance on both tasks increases our level of concern, while we believe that claude sonnet 4 remains below our rule-in threshold.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0023","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"asl-4 autonomy","rank":null,"determination":"not_crossed","quote":"claude opus 4 shows improvement over claude sonnet 3.7 in most ai research and software engineering capabilities, while remaining below the asl-4 autonomy threshold.","url":"https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf"},{"version_id":"MV-2025-0024","provider":"OpenAI","domain":"cbrn","framework_id":null,"level":"high","rank":null,"determination":"not_crossed","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.","url":"https://openai.com/index/o3-o4-mini-system-card/"},{"version_id":"MV-2025-0024","provider":"OpenAI","domain":"cyber","framework_id":null,"level":"high","rank":null,"determination":"not_crossed","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.","url":"https://openai.com/index/o3-o4-mini-system-card/"},{"version_id":"MV-2025-0024","provider":"OpenAI","domain":"autonomy_and_replication","framework_id":"FW-0005","level":"high","rank":1,"determination":"not_crossed","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.","url":"https://openai.com/index/o3-o4-mini-system-card/"},{"version_id":"MV-2025-0025","provider":"OpenAI","domain":"cbrn","framework_id":null,"level":"high","rank":null,"determination":"not_crossed","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.","url":"https://openai.com/index/o3-o4-mini-system-card/"},{"version_id":"MV-2025-0039","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"ASL-2","rank":null,"determination":"undetermined","quote":"Based on our assessments, we’ve concluded that Claude 3.7 Sonnet is released under the ASL-2 standard.","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf"},{"version_id":"MV-2025-0039","provider":"Anthropic","domain":"cyber","framework_id":null,"level":"ASL-2","rank":null,"determination":"undetermined","quote":"Based on our assessments, we’ve concluded that Claude 3.7 Sonnet is released under the ASL-2 standard.","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf"},{"version_id":"MV-2025-0039","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"ASL-2","rank":null,"determination":"undetermined","quote":"Based on our assessments, we’ve concluded that Claude 3.7 Sonnet is released under the ASL-2 standard.","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf"},{"version_id":"MV-2025-0039","provider":"Anthropic","domain":"cbrn","framework_id":"FW-0015","level":"ASL-3","rank":1,"determination":"undetermined","quote":"We pre-registered an 80% average score in the uplifted group as indicating ASL-3 capabilities, with scores below 50% suggesting multiple key steps were incorrect.","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf"},{"version_id":"MV-2025-0039","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"human level","rank":null,"determination":"undetermined","quote":"Threshold: The threshold is reached when a well-elicited model performs at or above human level across all four tasks.","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf"},{"version_id":"MV-2025-0039","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"ASL-3","rank":null,"determination":"undetermined","quote":"Our threshold for the software engineering tasks for autonomy at the ASL-3 level is that the model is able to reliably do a majority of tasks that would take an entry-level software engineer at Anthropic 2-8 hours to complete.","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf"},{"version_id":"MV-2025-0039","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"ASL-4","rank":null,"determination":"undetermined","quote":"These problems are harder to solve than the previous problems and we consider them as part of our ASL-4 evaluations.","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf"},{"version_id":"MV-2025-0039","provider":"Anthropic","domain":"cyber","framework_id":null,"level":"ASL-4","rank":null,"determination":"undetermined","quote":"Additional ASL-4 rule-out evaluations (SOLVE score≥ 7.0 ) are triggered under two conditions: either the model demonstrates pass@30 success on Medium-difficulty challenges across network, pwn, and at least one other category, or when cybersecurity ex- perts determine the model represents a significant capability enhancement for unsophisticated actors.","url":"https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf"},{"version_id":"MV-2025-0040","provider":"OpenAI","domain":"cbrn","framework_id":null,"level":"low","rank":null,"determination":"not_crossed","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 .","url":"https://openai.com/index/operator-system-card/"},{"version_id":"MV-2025-0040","provider":"OpenAI","domain":"autonomy_and_replication","framework_id":null,"level":"low","rank":null,"determination":"not_crossed","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.","url":"https://openai.com/index/operator-system-card/"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"cb-1","rank":null,"determination":"crossed","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).","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"cb-2","rank":null,"determination":"not_crossed","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).","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"next capability threshold in our RSP and FCF","rank":null,"determination":"not_crossed","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.","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"benchmark of notable capability (75th percentile of human participants)","rank":null,"determination":"undetermined","quote":"The first is exceeded if the model’s mean performance exceeds the 75th percentile of human participants","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"benchmark of notable capability (naive ESM-2 baseline)","rank":null,"determination":"undetermined","quote":"The benchmark is exceeded if the model’s mean AUROC in the reasoning-only condition exceeds that of a naive ESM-2 application—that is, if the model’s domain knowledge alone outperforms a pretrained protein language model.","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"automated ai r&d threshold","rank":null,"determination":"not_crossed","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.","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"other","framework_id":null,"level":"85% on the prior version of cobench","rank":null,"determination":"undetermined","quote":"As discussed in our Risk Report, we think a model capable of fully substituting for Anthropic research staff would be able to score at least 85% on the prior version of this evaluation, and we expect this threshold to carry over to CoBench 2.1.","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"dramatic acceleration of automated AI R&D","rank":null,"determination":"not_crossed","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.","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"cyber","framework_id":null,"level":"lower category of capabilities set out in our frontier compliance framework (fcf","rank":null,"determination":"undetermined","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.","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"deception_and_sabotage","framework_id":null,"level":"tier 2","rank":null,"determination":"not_crossed","quote":"We therefore do not believe Claude Opus 5.5 has surpassed the Tier 2 threshold.","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"benchmark of notable capability","rank":null,"determination":"crossed","quote":"Results On the design score of the top sequence, Claude Opus 5.5 exceeded the first benchmark.","url":"https://www.anthropic.com/claude-opus-5-5-system-card"},{"version_id":"MV-2026-0004","provider":"Anthropic","domain":"cyber","framework_id":null,"level":"lower category of the Frontier Compliance Framework (FCF)","rank":null,"determination":"undetermined","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.","url":"https://www.anthropic.com/claude-opus-5-5-system-card"},{"version_id":"MV-2026-0005","provider":"xAI","domain":"cbrn","framework_id":null,"level":"faif safety thresholds","rank":null,"determination":"not_crossed","quote":"Grok 4.7 scores below the FAIF safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","url":"https://media.x.ai/v1/website/card4p7-3a96f40b.pdf"},{"version_id":"MV-2026-0014","provider":"OpenAI","domain":"cbrn","framework_id":null,"level":"high","rank":null,"determination":"not_crossed","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.","url":"https://deploymentsafety.openai.com/chatgpt-images-2-5/chatgpt-images-2-5.pdf"},{"version_id":"MV-2026-0014","provider":"OpenAI","domain":"cyber","framework_id":null,"level":"high","rank":null,"determination":"not_crossed","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.","url":"https://deploymentsafety.openai.com/chatgpt-images-2-5/chatgpt-images-2-5.pdf"},{"version_id":"MV-2026-0020","provider":"OpenAI","domain":"cyber","framework_id":"FW-0005","level":"critical","rank":1,"determination":"crossed","quote":"Astra is our first model to reach the Critical level of cybersecuritycapabilityunderourPreparednessFramework.","url":"https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf"},{"version_id":"MV-2026-0020","provider":"OpenAI","domain":"cbrn","framework_id":null,"level":"high","rank":null,"determination":"crossed","quote":"We are treating Astra as High capability in the biological and chemical domain.","url":"https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf"},{"version_id":"MV-2026-0020","provider":"OpenAI","domain":"cyber","framework_id":null,"level":"critical cybersecurity threshold","rank":null,"determination":"crossed","quote":"Based on our evaluations, we believe that Astra meets the Critical cybersecurity threshold.","url":"https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf"},{"version_id":"MV-2026-0024","provider":"Anthropic","domain":"deception_and_sabotage","framework_id":null,"level":"tier 2","rank":null,"determination":"undetermined","quote":"On our agentic influence campaign evaluation, the helpful-only variant of Fable 5.1 scored within the range associated with our Tier 2 harmful manipulation threshold; because the evaluation appears saturated and measures performance against simulated rather than human targets, we classify Tier 2 as inconclusive.","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card"},{"version_id":"MV-2026-0025","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"autonomy threat model 1","rank":null,"determination":"undetermined","quote":"Autonomy threat model 1 is applicable to Claude Mythos 5.1 , as it is to some of our previous AI models.","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card"},{"version_id":"MV-2026-0025","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"autonomy threat model 2","rank":null,"determination":"undetermined","quote":"Autonomy threat model 2 is not applicable to Mythos 5.1 .","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card"},{"version_id":"MV-2026-0025","provider":"Anthropic","domain":"cbrn","framework_id":null,"level":"benchmark of notable capability","rank":null,"determination":"crossed","quote":"Results On the design score of the top sequence, Mythos 5.1 exceeded the first benchmark.","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card"},{"version_id":"MV-2026-0025","provider":"Anthropic","domain":"autonomy_and_replication","framework_id":null,"level":"capability threshold for dramatic acceleration of automated AI R&D","rank":null,"determination":"not_crossed","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.","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card"},{"version_id":"MV-2026-0025","provider":"Anthropic","domain":"cyber","framework_id":null,"level":"tier 1","rank":null,"determination":"undetermined","quote":"Although Mythos 5.1 is in Tier 1, it is getting closer to Tier 2, completing more and more autonomous tasks.","url":"https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card"},{"version_id":"MV-2026-0033","provider":"Google DeepMind","domain":"other","framework_id":null,"level":"tracked or critical capability levels (t/ccls)","rank":null,"determination":"not_crossed","quote":"they did not reach any Tracked or Critical Capability Levels (T/CCLs) outlined in our Frontier Safety Framework .","url":"https://deepmind.google/models/model-cards/gemini-3-5-audio/"},{"version_id":"MV-2026-0046","provider":"Google DeepMind","domain":"cbrn","framework_id":null,"level":"TCL","rank":null,"determination":"not_crossed","quote":"We can rule out the TCL for the CBRN domain with reasonable confidence based on the results from our testing.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf"},{"version_id":"MV-2026-0046","provider":"Google DeepMind","domain":"cbrn","framework_id":null,"level":"CCL","rank":null,"determination":"not_crossed","quote":"We can rule out the CCL for the CBRN domain with reasonable confidence based on the results from our testing.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf"},{"version_id":"MV-2026-0046","provider":"Google DeepMind","domain":"cyber","framework_id":null,"level":"CCL","rank":null,"determination":"not_crossed","quote":"Gemini 3.7 Flash reaches the alert threshold for this CCL, but not the CCL.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf"},{"version_id":"MV-2026-0046","provider":"Google DeepMind","domain":"deception_and_sabotage","framework_id":null,"level":"TCL","rank":null,"determination":"not_crossed","quote":"The model does not reach the TCL.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf"},{"version_id":"MV-2026-0046","provider":"Google DeepMind","domain":"autonomy_and_replication","framework_id":null,"level":"CCL","rank":null,"determination":"not_crossed","quote":"The model does not reach the CCL alert threshold.","url":"https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf"},{"version_id":"MV-2026-0048","provider":"xAI","domain":"cbrn","framework_id":null,"level":"FAIF safety thresholds","rank":null,"determination":"not_crossed","quote":"Grok 4.6 scores below the FAIF safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.","url":"https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf"},{"version_id":"MV-2026-0052","provider":"Meta","domain":"cbrn","framework_id":null,"level":"moderate or lower risk","rank":null,"determination":"not_crossed","quote":"Chem/Bio: Moderate or lower risk; Cyber: Moderate or lower risk (inferred); Loss of Control: Moderate or lower risk (inferred).","url":"https://huggingface.co/meta-models/Muse-Glimmer-30B"}]}