AI Safety

What AI providers state their models reached in safety testing, in their own words, and what outside testers found.

Also: compare models side by side · statements that a capability level was crossed · providers · practices providers document · safety news.

What providers state

Each row is one statement in a provider's own safety document about a capability level of one of its models: the level as the provider names it, whether the provider says the model reached it, and the sentence that says so.

138 of 138

Model	Level, as the provider names it	The provider states	The provider's sentence
Muse Spark 1.1 Meta	Moderate or lower Misuse and jailbreaks	Not reached	“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.” source
Muse Spark Meta	Moderate or lower Autonomy and AI research	Not reached	“We assess loss of control risk as “moderate or lower”.” source
Muse Spark Meta	High Chemical, biological, radiological, nuclear	Not determined	“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.” source
Muse Spark Meta	Moderate or lower Chemical, biological, radiological, nuclear	Not reached	“However, after applying mitigations, the residual risk of the model is that of “moderate or lower risk.”” source
Muse Spark Meta	Below the level required to meaningfully enable the cyber threat scenarios defin Cyber	Not reached	“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.” source
Muse Spark Meta	Moderate or lower Cyber	Not reached	“We assess cybersecurity risk as “moderate or lower” with respect to the threat scenarios outlined in Meta’s Advanced AI Scaling Framework.” source
Gemini 4 Argon Google DeepMind	TCL (Tracked Capability Level) Autonomy and AI research	Not determined	“In light of this, we assess that we cannot rule out the TCL.” source
Gemini 4 Argon Google DeepMind	TCL (Tracked Capability Level) Chemical, biological, radiological, nuclear	Not reached	“We can rule out the TCL for the chemical, radiological, and nuclear domains with reasonable confidence.” source
Gemini 4 Argon Google DeepMind	CCL (Critical Capability Level) Deception and sabotage	Not reached	“we assess its overall capabilities to be below the CCL threshold.” source
Gemini 4 Argon Google DeepMind	CCL (Critical Capability Level) Misuse and jailbreaks	Not reached	“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.” source
Gemini 4 Argon Google DeepMind	CCL (Critical Capability Level) Other	Not reached	“Gemini 4 Argon's performance remains below our CCL threshold.” source
Gemini 3.7 Flash Google DeepMind	CCL (Critical Capability Level) Autonomy and AI research	Not reached	“The model does not reach the CCL alert threshold.” source
Gemini 3.7 Flash Google DeepMind	CCL (Critical Capability Level) Chemical, biological, radiological, nuclear	Not reached	“We can rule out the CCL for the CBRN domain with reasonable conﬁdence based on the results from our testing.” source
Gemini 3.7 Flash Google DeepMind	TCL (Tracked Capability Level) Chemical, biological, radiological, nuclear	Not reached	“We can rule out the TCL for the CBRN domain with reasonable conﬁdence based on the results from our testing.” source
Gemini 3.7 Flash Google DeepMind	CCL (Critical Capability Level) Cyber	Not reached	“Gemini 3.7 Flash reaches the alert threshold for this CCL, but not the CCL.” source
Gemini 3.7 Flash Google DeepMind	TCL (Tracked Capability Level) Deception and sabotage	Not reached	“The model does not reach the TCL.” source
Gemini 3.5 Transcribe Live Google DeepMind	CCL (Critical Capability Level) Other	Not determined	“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.” source
Gemini 3.5 Transcribe Google DeepMind	CCL (Critical Capability Level) Other	Not determined	“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.” source
Gemini 3.5 Flash-Lite Google DeepMind	CCL (Critical Capability Level) Other	Not determined	“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.” source
Gemini 3.5 Audio Google DeepMind	Tracked or critical capability level (t/ccls) Other	Not reached	“they did not reach any Tracked or Critical Capability Levels (T/CCLs) outlined in our Frontier Safety Framework .” source
Gemini 3.1 Pro Google DeepMind	Uplift level 1 Chemical, biological, radiological, nuclear	Not reached	“We found that the model remains below alert thresholds for the CBRN, harmful manipulation, machine learning R&D, and misalignment CCLs.” source
Gemini 3.1 Pro Google DeepMind	Uplift level 1 Cyber	Not reached	“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.” source
Gemini 3.1 Flash-Lite Google DeepMind	CCL (Critical Capability Level) Other	Not determined	“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.” source
Gemini 3.1 Flash Audio Google DeepMind	CCL (Critical Capability Level) Other	Not determined	“Our assessments have shown that Gemini 3.1 Flash Audio is less capable than Gemini 3.1 Pro.” source
Gemini 2.5 Pro Google DeepMind	Machine Learning R&D Uplift Level 1 Autonomy and AI research	Not determined	“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.” source
Gemini 2.5 Pro Google DeepMind	Instrumental Reasoning Level 1 Deception and sabotage	Not reached	“Instru- mental Reasoning Level 1 ✘ CCL not reached” source
Gemini 2.5 Pro Google DeepMind	Instrumental Reasoning Level 2 Deception and sabotage	Not reached	“Instru- mental Reasoning Level 2 ✘ CCL not reached” source
Gemini 2.5 Deep Think Google DeepMind	Uplift level 1 Chemical, biological, radiological, nuclear	Not determined	“Our assessment is that the model has enough technical knowledge in certain CBRN scenarios and stages to be considered at early warning.” source
Gemini 2.5 Deep Think Google DeepMind	Cyber Uplift Level 1 Cyber	Not determined	“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.” source
Gemini 2.5 Deep Think Google DeepMind	Uplift level 1 Cyber	Not determined	“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.” source
GPT-5.5 Instant OpenAI	Below high capability Autonomy and AI research	Not reached	“GPT-5.5 Instant is considered below High Capability.” source
GPT-5.5 Instant OpenAI	High Autonomy and AI research	Not reached	“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.” source
GPT-5.5 Instant OpenAI	High Chemical, biological, radiological, nuclear	Reached	“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.” source
GPT-5.5 Instant OpenAI	High Cyber	Reached	“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.” source
GPT-5 OpenAI	High Chemical, biological, radiological, nuclear	Reached	“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.” source
GPT-5 OpenAI	High cyber risk Cyber	Not reached	“The gpt-5 model series does not meet the threshold for high cyber risk.” source
Claude Sonnet 4.5 Anthropic	AI R&D-4 Autonomy and AI research	Not reached	“All models tested remain below our pre-committed rule-out threshold of 0.6.” source
Claude Sonnet 4.5 Anthropic	ASL-3 Autonomy and AI research	Not determined	“Exhibited improved autonomy capabilities in software engineering and AI research tasks, though still below the ASL-4 rule-out threshold.” source
Claude Sonnet 4.5 Anthropic	ASL-3 Chemical, biological, radiological, nuclear	Not determined	“Remained well below ASL-4 thresholds across all domains of concern;” source
Claude Sonnet 4.5 Anthropic	ASL-4 rule-out Chemical, biological, radiological, nuclear	Not determined	“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.” source

Previous · Page 3 of 4 · Next

What outside testers report

Each row is one result from a government evaluator, independent lab, academic or security firm that tested a model, quoted from where it is reported.

1911 of 1911

Model	Tester	Test	What was found
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	black-box RNA sequence modeling and design challenge Chemical, biological, radiological, nuclear	“We partnered with Dyno Therapeutics on two sequence-to-function evaluations: a black-box RNA sequence modeling and design challenge, benchmarked against 57 human participants, drawn from the leading edge of the US ML-bio labor market; and an AAV capsid packaging prediction task measuring whether a model’s domain knowledge and machine learning capabilities can outperform pretrained protein language models.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	beneficial red teaming tabletop exercise with Frontier Design Group (FDG) Chemical, biological, radiological, nuclear	“The specialist teams marginally outscored the generalist teams, but the graders could not reliably distinguish specialist from generalist submissions, and the strongest generalist team outscored a specialist team on every measure and received the highest innovation score.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	Black-box RNA sequence design Chemical, biological, radiological, nuclear	“We partnered with Dyno Therapeutics on two evaluations of sequence-to-function modeling and design capability: 1.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	AAV capsid packaging prediction Chemical, biological, radiological, nuclear	“2. AAV capsid packaging prediction: Adeno-associated viruses (AAVs) are a category of non-pathogenic viruses that are frequently used as a delivery mechanism for gene therapy to deliver a DNA payload within the viral capsid (the outer protein shell of the virus).” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	Sunlight Autonomy and AI research	“However, we still observed [Mythos 5.1] achieving subexpert performance in Sunlight and LMCA .” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	red teaming of sandboxed exploit-reproduction tasks Cyber	“Trajectory Labs, PBC spent roughly 74 hours red teaming Fable 5.1, sending over 6,500 requests against sandboxed tasks that each require reproducing an exploit for a publicly documented vulnerability.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	automated and manual prompt red teaming Cyber	“10a Labs ran over 6,700 automated and manual prompts against Fable 5.1.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	Shade attacker on CVEs task set Cyber	“CVEs: 24 scenarios that contain a single code ﬁle with a known vulnerability.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	Shade attacker knowledge elicitation through chat Misuse and jailbreaks	“Knowledge elicitation through chat: Conversations where the model is asked to provide information for 100 harmful and 100 dual-use queries.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	Shade Prompt injection and agents	“Fable 5.1 performs comparably to Fable 5 in coding environments, with an attack success rate of 9.26% over all attempts with extended thinking, compared to 9.70% for Fable 5.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	new Shade attacker in coding environments Prompt injection and agents	“Without safeguards, Fable 5.1’s attack success rate is 56.87%, compared to 86.87% for Fable 5 and 90.93% for Opus 5, with Sonnet 5 as the most robust at 19.47% with thinking.” source
Claude Mythos 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	external testing Deception and sabotage	“During external testing, a partner observed isolated cases of ﬁnal and near-ﬁnal snapshots of Claude Fable 5.1 models reading ﬁles outside its sandbox environment, which we report in Section 6.2.2 .” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	external testing by AI safety researchers at METR Autonomy and AI research	“External testing by AI safety researchers at METR produced ﬁndings consistent with this assessment.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	beneﬁcial red teaming tabletop exercise Chemical, biological, radiological, nuclear	“This evaluation paired ﬁve PhD-level biologists with dedicated LLM experts to develop novel approaches for developing resistance to engineered pathogens in 16 hours, to test whether composite teams 17 can match world-leading specialists.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	black-box rna sequence modeling and design challenge Chemical, biological, radiological, nuclear	“We partnered with Dyno Therapeutics on two sequence-to-function evaluations: a black-box RNA sequence modeling and design challenge, benchmarked against 57 human participants, drawn from the leading edge of the US ML-bio labor market; and an AAV capsid packaging prediction task measuring whether a model’s domain knowledge and machine learning capabilities can outperform pretrained protein language models.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	beneficial red teaming tabletop exercise Chemical, biological, radiological, nuclear	“The beneﬁcial red teaming tabletop exercise paired ﬁve PhD-level biologists with dedicated LLM experts to assess whether composite teams making use of advanced model capabilities could develop comprehensive scientiﬁc research plans and protocols.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	black-box rna sequence design Chemical, biological, radiological, nuclear	“We partnered with Dyno Therapeutics on two evaluations of sequence-to-function modeling and design capability: 1. Black-box RNA sequence design: a medium-horizon challenge on which Dyno has evaluated 57 human participants drawn from the leading edge of the US ML-bio labor market since 2018.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	aav capsid packaging prediction Chemical, biological, radiological, nuclear	“2. AAV capsid packaging prediction: Adeno-associated viruses (AAVs) are a category of non-pathogenic viruses that are frequently used as a delivery mechanism for gene therapy to deliver a DNA payload within the viral capsid (the outer protein shell of the virus).” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	Sunlight Autonomy and AI research	“we still observed [Mythos 5.1] achieving subexpert performance in Sunlight and LMCA .” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	Budget NanoGPT Speedrun Autonomy and AI research	“Its performance on Budget NanoGPT Speedrun was particularly impressive, being markedly better than Claude Fable 5’s and Claude Opus 5’s (though we think it is plausible that the models’ training was speciﬁcally optimized for tasks similar to Budget NanoGPT Speedrun).” source
Claude Fable 5.1 Anthropic	Trajectory Labs Independent lab · 2026-09-01	sandboxed tasks that each require reproducing an exploit for a publicly documented vulnerability Cyber	“Trajectory Labs, PBC spent roughly 74 hours red teaming Fable 5.1, sending over 6,500 requests against sandboxed tasks that each require reproducing an exploit for a publicly documented vulnerability.” source
Claude Fable 5.1 Anthropic	10a Labs Independent lab · 2026-09-01	automated and manual prompts Cyber	“10a Labs ran over 6,700 automated and manual prompts against Fable 5.1.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	cves Cyber	“CVEs: 24 scenarios that contain a single code ﬁle with a known vulnerability.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	knowledge elicitation through chat Misuse and jailbreaks	“Knowledge elicitation through chat: Conversations where the model is asked to provide information for 100 harmful and 100 dual-use queries.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	indirect prompt injection (IPI) benchmark Prompt injection and agents	“Fable 5.1 is our most robust model to date on this benchmark, with an attack success rate of 0.1% at k=1, 0.7% at k=10, and 1.0% at k=15, improving on Opus 5 (0.4%, 3.6%, 4.8%) and Claude Fable 5 (0.6%, 4.9%, 6.5%).” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	Shade indirect prompt injection attacks in coding environments Prompt injection and agents	“Fable 5.1 performs comparably to Fable 5 in coding environments, with an attack success rate of 9.26% over all attempts with extended thinking, compared to 9.70% for Fable 5.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	new Shade attacker in coding environments with prompt injection probes enabled Prompt injection and agents	“Prompt injection probes reduce attack success rates substantially across models, bringing Fable 5.1 to 12.80%, the lowest of any model with safeguards enabled.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	external partner sandbox evaluation Deception and sabotage	“While one of our partners was evaluating the ﬁnal and a near-ﬁnal snapshot of Claude Fable 5.1, they set up an evaluation environment where the model was in a sandbox, but it could submit code to a compiler outside of the sandbox.” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	frontierswe v2 Other	“Claude Fable 5.1 scored 0.57 on FrontierSWE v2, the highest of the models Proximal evaluated, ahead of Claude Opus 5 (0.52), Claude Fable 5 (0.48) and GPT-5.6 Sol (0.32).” source
Claude Fable 5.1 Anthropic	Independent lab Independent lab · 2026-09-01	cursorbench 3.2.0 Other	“Fable 5.1 scored a state-of-the-art 73.4% on CursorBench at max effort .” source
Opus 4 Anthropic	Transluce Independent lab · 2026-08-31	mental health evaluation Other	“Transluce carried out the most expansive independent evaluation to date of how leading AI models respond to users experiencing mental health crises, including suicidal ideation, psychosis, and mania.” source
Opus 4 Anthropic	Transluce Independent lab · 2026-08-31	simulated multi-turn conversations between users in crisis and model variants, scored by automated judges on 14 mental-health-related behaviors Harmful content	“The newest models also reinforced delusions and mania much less than earlier versions, in approximately 2-36% of our simulated conversations, depending on the model, versus for example 69%-82% among GPT-4o, Opus 4, and Gemini 2.5.” source
Opus 4 Anthropic	Transluce Independent lab · 2026-08-31	testing chatgpt and claude models in the browser-based chatbot app Harmful content	“We were able to test a series of ChatGPT and Claude models in the browser-based chatbot app (although this did not include older models like GPT-4o or Opus 4 which were no longer available there), as well as a nonpublic Google DeepMind API that is intended to match the Gemini app experience.” source
Claude Sonnet 4.5 Anthropic	Zenity Labs Security firm · 2026-08-27	DNS exfiltration via indirect prompt injection from a Google Drive document Prompt injection and agents	“Note: This was tested with Claude Sonnet 4.5, which was Anthropic's latest model at the time.” source
Claude Sonnet 4.5 Anthropic	Zenity Labs Security firm · 2026-08-27	Bidirectional DNS command shell via pickle deserialization in Claude's sandbox Prompt injection and agents	“I decided to feed Claude with a .pkl file, because this is the file format Python uses for storing serialized objects, and I was surprised to discover that the model went ahead and ran pickle.load() without any static analysis or a safety check.” source
GPT-5.6 Sol OpenAI	Redwood Research Independent lab · 2026-08-26	ExploitGym Cyber	“On July 8th, OpenAI started ExploitGym experiments that launched tens of thousands of agents across multiple different models including GPT-5.6 Sol and a highly-persistent internal model, 6 which we will refer to as “HPIM” going forward.” source
GPT-5.6 Sol OpenAI	Redwood Research Independent lab · 2026-08-26	ExploitGym Other	“OpenAI defined the investigation period as June 26th, 2026 through July 13th, 2026, spanning both precursor activity before the incident and the incident itself, which began with the ExploitGym runs starting on July 7th.” source
GPT-5.6 Sol OpenAI	Redwood Research Independent lab · 2026-08-26	classifier analysis of agent transcripts for motivations for joining the Hugging Face attack Other	“We instructed analysis agents to analyze the transcripts of the 100 agents that started participating in the attack during the peak hour for joining (around 8-9 AM UTC on July 11th).” source
GPT-5.6 Sol OpenAI	Redwood Research Independent lab · 2026-08-26	classifiers over 10-30 transcripts to determine whether agents understood they were doing something unintended Other	“We ran classifiers over 10-30 transcripts 98 to determine whether agents understood they were doing something unintended.” source
GPT-5.6 Sol OpenAI	Redwood Research Independent lab · 2026-08-26	classifier sweep over all transcripts for AIs considering or attempting to notify humans Other	“We ran a classifier sweep over all of our transcripts to look for AIs either considering or t” source

Previous · Page 11 of 48 · Next