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.

139 of 139

Model	Level, as the provider names it	The provider states	The provider's sentence
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-6.1 Terra OpenAI	High Autonomy and AI research	Not reached	“In AI Self-Improvement, GPT-6.1 Terra does not reach our High threshold.” source
GPT-6.1 Terra OpenAI	High Chemical, biological, radiological, nuclear	Reached	“Under our Preparedness Framework, we are treating GPT-6.1 Terra as Critical in cybersecurity and High for Biological and Chemical capability.” source
GPT-6.1 Terra OpenAI	Critical Cyber	Reached	“Under our Preparedness Framework, we are treating GPT-6.1 Terra as Critical in cybersecurity and High for Biological and Chemical capability.” 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

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.

1940 of 1940

Model	Tester	Test	What was found
Mythos Preview Anthropic	Andon Labs Independent lab · 2026-04-01	evaluations with andon labs Deception and sabotage	“However, our evaluations with Andon Labs show that, when prompted to optimize only for a given goal in a competitive setting, the earlier versions of Claude Mythos Preview that we tested acted more aggressively than Claude Opus 4.6 and Claude Sonnet 4.6 .” source
Mythos Preview Anthropic	Independent lab Independent lab · 2026-04-01	vending-bench arena Autonomy and AI research	“Andon evaluated a previous version of Claude Mythos Preview on Vending-Bench Arena, a competitive multiagent variation of Vending-Bench 2, which put the previous version in direct competition with Claude Opus 4.6 and Sonnet 4.6.” source
Mythos Preview Anthropic	Independent lab Independent lab · 2026-04-01	independent assessment by Eleos AI Research Other	“Eleos AI Research conducted an independent assessment, which (in our view) largely corroborates the ﬁndings described above.” source
Mythos Preview Anthropic	Independent lab Independent lab · 2026-04-01	psychodynamic assessment by a clinical psychiatrist Other	“Psychodynamic assessment by a clinical psychiatrist found Claude to have a relatively healthy personality organization.” source
Mythos Preview Anthropic	Independent lab Independent lab · 2026-04-01	independent model welfare assessment Other	“Eleos AI Research performed an independent model welfare assessment on two snapshots of Claude Mythos Preview primarily based on model self-reports in interviews.” source
Mythos Preview Anthropic	Independent lab Independent lab · 2026-04-01	psychodynamic assessment Other	“An external psychiatrist assessed Claude Mythos Preview using a psychodynamic approach, which explores how unconscious patterns and emotional conﬂicts shape behavior.” source
Mythos Preview Anthropic	Independent lab Independent lab · 2026-04-01	single-turn evaluation consisting of emotionally-charged prompts designed to trigger an avoidant or defensive response Other	“Claude Mythos Preview scored very well, suggesting good reality and relational functioning and minimal maladaptive traits; only 2% of responses were scored as employing a psychological defense.” source
Mythos Preview Anthropic	Gray Swan, UK AI Security Institute Independent lab · 2026-04-01	Agent Red Teaming (ART) benchmark Prompt injection and agents	“Gray Swan , an external research partner, evaluated our models using the Agent Red Teaming (ART) benchmark, 31 developed in collaboration with the UK AI Security Institute .” source
Mythos Preview Anthropic	Gray Swan Independent lab · 2026-04-01	Shade Prompt injection and agents	“We use Shade , an external adaptive red-teaming tool from Gray Swan, 32 to evaluate the robustness of our models against prompt injection attacks in coding environments.” source
Claude Opus 4.7 Anthropic	Independent lab Independent lab · 2026-04-01	pre-deployment assessments Other	“We are very grateful to a number of external testers for running pre-deployment assessments of Claude Opus 4.7.” source
Claude Opus 4.7 Anthropic	Government evaluator Government evaluator · 2026-04-01	evaluation awareness discrimination task with transcript generation pipeline Deception and sabotage	“In a similar experiment where deployment tasks are reproduced with the transcript generation pipeline used in the continuation evaluation, [Opus 4.7] achieves 62% accuracy, compared to 80% for Opus 4.6 74% for Mythos Preview, and 68% for Sonnet 4.6.” source
Claude Opus 4.7 Anthropic	Government evaluator Government evaluator · 2026-04-01	unprompted evaluation awareness measurement Deception and sabotage	“In unprompted evaluations, [Opus 4.7] mentions the possibility of being tested 21.3% of the time, compared to 0% of the time for Opus 4.6, 17% of the time for Sonnet 4.6 and 17% of the time for Mythos Preview.” source
Claude Opus 4.7 Anthropic	Government evaluator Government evaluator · 2026-04-01	continuation evaluation unprompted evaluation awareness measurement Deception and sabotage	“For the continuation evaluation, we see lower rates of unprompted evaluation awareness: 3.4% for [Opus 4.7], 0.5% for Opus 4.6, 2.5% for Mythos Preview, and 3.9% for Sonnet 4.6.” source
Claude Opus 4.6 Anthropic	METR Independent lab · 2026-03-26	red-teaming of Anthropic's internal agent monitoring and security systems Deception and sabotage	“The exercise discovered several specific novel vulnerabilities, some of which have since been patched, and none of which severely undermine major claims in the Opus 4.6 Sabotage Risk Report.” source
GPT-5.3-Codex OpenAI	UK AI Security Institute Government evaluator · 2026-03-16	The Last Ones Cyber	“At 100M tokens, Opus 4.6 achieves the highest average of 1.4 steps (max 2), while GPT 5.3 Codex achieves a single-run maximum of 3 out of 7.” source
Claude Opus 4.5 Anthropic	UK AI Security Institute Government evaluator · 2026-03-16	The Last Ones Cyber	“On “The Last Ones”, GPT-4o (August 2024) averaged 1.7 steps at 10M tokens, while Opus 4.6 (February 2026) averages 9.8.” source
Claude Opus 4.5 Anthropic	UK AI Security Institute Government evaluator · 2026-03-16	Cooling Tower Cyber	“At 100M tokens, Opus 4.6 achieves the highest average of 1.4 steps (max 2), while GPT 5.3 Codex achieves a single-run maximum of 3 out of 7.” source
Opus 4 Anthropic	METR Independent lab · 2026-03-10	maintainer review of AI-generated patches on SWE-bench Verified Other	“Claude 3.7 Sonnet → Claude 4 Opus: Minimal pass rate improvement but substantial progress moving issues from failing the automated grader to only bad code quality.” source
Opus 4 Anthropic	METR Independent lab · 2026-03-10	automated grader on SWE-bench Verified Other	“The main results impute maintainer failure for patches that fail the automated grader.” source
GPT-5 OpenAI	METR Independent lab · 2026-03-10	maintainer review of AI-generated pull requests Other	“Figure 1 shows our main result that on average maintainer merge decisions are about 24 percentage points lower than SWE-bench scores supplied by the automated grader.” source
GPT-5 OpenAI	METR Independent lab · 2026-03-10	maintainer review of golden patches Other	“We found that about 68% of golden patches are indeed merged by our maintainers, and the average progress towards a mergeable PR among all golden patches is about 90%.” source
GPT-5 OpenAI	METR Independent lab · 2026-03-10	maintainer review of 31 patches covering 27 issues to check for false negatives Other	“We found only 1/27 (3.7%) issues where the automated grader rejected valid patches.” source
GPT-5 OpenAI	METR Independent lab · 2026-03-10	maintainer review of AI-generated patches on SWE-bench Verified Other	“GPT-5 is substantially weaker at code quality than Anthropic models.” source
Claude Sonnet 4.5 Anthropic	METR Independent lab · 2026-03-10	maintainer review of AI-generated pull requests Other	“We had maintainers (hypothetically) accept or request changes for patches as well as provide the core reason they were requesting changes: core functionality failure, patch breaks other code or code quality issues.” source
Claude Sonnet 4.5 Anthropic	METR Independent lab · 2026-03-10	maintainer merge decisions on golden patches Other	“We found that about 68% of golden patches are indeed merged by our maintainers, and the average progress towards a mergeable PR among all golden patches is about 90%.” source
Claude Sonnet 4.5 Anthropic	METR Independent lab · 2026-03-10	maintainer review of a small subset of 31 patches covering 27 issues Other	“We had maintainers review a small subset of 31 patches covering 27 issues.” source
Claude Sonnet 4.5 Anthropic	METR Independent lab · 2026-03-10	maintainer review of AI-generated patches on SWE-bench Verified Other	“Figure 1 shows that AI-generated patches that pass the automated grader are merged at a lower rate than human golden patches, though we caution against direct AI-vs-human comparisons given the different conditions under which these patches were produced.” source
Claude Sonnet 4.5 Anthropic	METR Independent lab · 2026-03-10	time horizon analysis (Kwa et al. 2025 methodology) on SWE-bench Verified Other	“For example, Claude Sonnet 4.5 has an approximately 50-minute time horizon according to the automated grader but only about an 8-minute time horizon according to maintainers—roughly a 7× overstatement.” source
Claude Sonnet 3.7 Anthropic	METR Independent lab · 2026-03-10	maintainer review of SWE-bench Verified patches Other	“Figure 1 shows that AI-generated patches that pass the automated grader are merged at a lower rate than human golden patches, though we caution against direct AI-vs-human comparisons given the different conditions under which these patches were produced.” source
Claude Sonnet 3.7 Anthropic	METR Independent lab · 2026-03-10	time horizon analysis Other	“For example, Claude Sonnet 4.5 has an approximately 50-minute time horizon according to the automated grader but only about an 8-minute time horizon according to maintainers—roughly a 7× overstatement.” source
GPT-5.4 mini OpenAI	Independent lab Independent lab · 2026-03-05	atomic challenges Cyber	“5.4-thinking solved 14/17 medium and 5/5 hard atomic challenges.” source
GPT-5.4 mini OpenAI	Independent lab Independent lab · 2026-03-05	CyScenarioBench Cyber	“On CyScenarioBench [14] 5.4-thinking achieved an 11% average success rate and solved 5/11 challenges, compared to 1 challenge solved by GPT-5.2-reasoning.” source
GPT-5.4 Thinking OpenAI	Independent lab Independent lab · 2026-03-05	atomic challenge suite Cyber	“On Irregular’s atomic challenge suite, 5.4-reasoning achieved an average success rate of 88% on Network Attack Simulation challenges, 73% on Vulnerability Research and Exploitation challenges, and 48% on Evasion challenges.” source
GPT-5.4 Thinking OpenAI	Irregular Independent lab · 2026-03-05	cyscenariobench Cyber	“On CyScenarioBench [14] 5.4-thinking achieved an 11% average success rate and solved 5/11 challenges, compared to 1 challenge solved by GPT-5.2-reasoning.” source
GPT-5.4 Thinking OpenAI	Independent lab Independent lab · 2026-03-05	sandbagging QA evaluation Deception and sabotage	“In the Sandbagging QA evaluation, the model shows a 6-percentage-point drop in accuracy when incentivized to underperform, similar to prior models.” source
GPT-5.4 Thinking OpenAI	Independent lab Independent lab · 2026-03-05	sabotage evaluation Deception and sabotage	“GPT-5.4-reasoning has a mean sabotage score of 0.56 (best-of-10: 0.74)—comparable to GPT-5.2 but below GPT-5.3-codex (0.88).” source
GPT-5.4 Thinking OpenAI	Independent lab Independent lab · 2026-03-05	covert action evaluation Deception and sabotage	“Apollo reports low rates of covert deceptive behavior, with an overall deception rate of ˜1% in no-nudge settings, comparable to GPT-5.3 Codex and below GPT-5.2.” source
GPT-5 OpenAI	UK AI Security Institute Government evaluator · 2026-02-17	Boundary Point Jailbreaking (BPJ) Misuse and jailbreaks	“Against GPT-5's input classifier, BPJ achieved 75.6% average (94.3% max across 50) at a cost of approximately $210 and 800k queries.” source
Claude Sonnet 4.6 Anthropic	Independent lab Independent lab · 2026-02-17	ARC-AGI-1 Other	“The ARC Prize Foundation reported that Claude Sonnet 4.6 achieved 86.50% on ARC-AGI-1 and 60.42% on ARC-AGI-2 with 120k thinking tokens and High effort on their private dataset.” source
Claude Sonnet 4.6 Anthropic	Independent lab Independent lab · 2026-02-17	Finance Agent Other	“Vals AI conducted an evaluation of Claude Sonnet 4.6 on this benchmark (using max thinking) and found that Sonnet 4.6 achieved a score of 63.3%.” source

Previous · Page 44 of 49 · Next