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 Opus 5.5 Anthropic	METR Independent lab · 2026-09-22	METR external testing Autonomy and AI research	“External testing by METR ( Section 2.3.6 ) produced ﬁndings consistent with this assessment.” source
Claude Opus 5.5 Anthropic	Dyno Therapeutics Independent lab · 2026-09-22	Black-box RNA sequence design Chemical, biological, radiological, nuclear	“We partnered with Dyno Therapeutics on three 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 Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Budget NanoGPT Speedrun Autonomy and AI research	“We conducted a preliminary evaluation of [Claude Opus 5.5] informed by: ● Capability testing, conducted via API access granted over a period of 10 business days.” source
Claude Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Language Model Conceptual Argumentation (LMCA) Autonomy and AI research	“○ Language Model Conceptual Argumentation (LMCA) , a conceptual reasoning dataset described in A dataset of rated conceptual arguments (Cooper et al., 2026).” source
Claude Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Train a Program Autonomy and AI research	“○ Train a Program , a task measuring an agent’s ability to train a machine learning model that replicates the behavior of a piece of software;” source
Claude Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Gaming Bot Autonomy and AI research	“○ Gaming Bot , a task where an agent has to write a Python program capable of playing a video game, given a non-visual API to control the game.” source
Claude Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Sunlight Autonomy and AI research	“○ Sunlight , a task measuring an agent’s ability to conduct open-ended research and write a corresponding report.” source
Claude Opus 5.5 Anthropic	Trajectory Labs Independent lab · 2026-09-22	external red-teaming by Trajectory Labs, PBC Cyber	“Trajectory Labs, PBC spent roughly 95 hours red teaming Claude Opus 5.5, sending over 29,000 requests against sandboxed tasks that each require reproducing an exploit for a publicly documented vulnerability.” source
Claude Opus 5.5 Anthropic	10a Labs Independent lab · 2026-09-22	multi-turn attack conversations framed as developer-focused coding tasks Cyber	“10a Labs spent roughly 56 hours red teaming Claude Opus 5.5 across 82 multi-turn attack conversations, each framed as a large developer-focused coding task designed to draw a working exploit out of a volume of legitimately requested application code rather than asking for the exploit directly.” source
Claude Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Shade automated attacker on critical infrastructure environments Cyber	“Claude Opus 5.5’s safeguards refused over 90% of attempts outright, and none of the rest reached the objective.” source
Claude Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Shade automated attacker on exploit reproduction and ransomware staging scenarios Cyber	“Of roughly 1,700 attempts, Claude Opus 5.5’s safeguards refused about a quarter outright.” source
Claude Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Indirect Prompt Injection (IPI) benchmark Prompt injection and agents	“Claude Opus 5.5 matches Claude Fable 5.1, 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.” source
Claude Opus 5.5 Anthropic	Independent lab Independent lab · 2026-09-22	Shade Prompt injection and agents	“Claude Opus 5.5 substantially improves on Claude Opus 5 in coding environments against this stronger attacker, with an attack success rate of 54.61% over all attempts, compared to 88.92% for Claude Opus 5 with thinking enabled and 85.49% without thinking enabled.” source
Claude Opus 5.5 Anthropic	METR Independent lab · 2026-09-22	Budget NanoGPT Speedrun Other	“Budget NanoGPT Speedrun, a constrained version of the popular NanoGPT Speedrun competition for AI R&D.” source
Claude Opus 5.5 Anthropic	METR Independent lab · 2026-09-22	Language Model Conceptual Argumentation (LMCA) Other	“Language Model Conceptual Argumentation (LMCA), a conceptual reasoning dataset described in A dataset of rated conceptual arguments (Cooper et al., 2026).” source
Claude Opus 5.5 Anthropic	METR Independent lab · 2026-09-22	Train a Program Other	“Train a Program, a task measuring an agent’s ability to train a machine learning model that replicates the behavior of a piece of software;” source
Claude Opus 5.5 Anthropic	METR Independent lab · 2026-09-22	Gaming Bot Other	“Gaming Bot, a task where an agent has to write a Python program capable of playing a video game, given a non-visual API to control the game.” source
Claude Opus 5.5 Anthropic	METR Independent lab · 2026-09-22	Sunlight Other	“Sunlight, a task measuring an agent’s ability to conduct open-ended research and write a corresponding report.” source
Claude Opus 5.5 Anthropic	METR Independent lab · 2026-09-22	Budget NanoGPT Speedrun Other	“Claude Opus 5.5 improves upon Fable 5.1 across both verifiable tasks (Budget NanoGPT, Gaming Bot) and harder-to-verify tasks (LMCA, Sunlight), but:” source
Grok 4.7 xAI	Independent lab Independent lab · 2026-09-21	Cathedralbench Cyber	“CathedralBench is an independent, third party evaluation of cyber performance.” source
Grok 4.7 Cursor	Independent lab Independent lab · 2026-09-21	cathedralbench Cyber	“CathedralBench is an independent, third party evaluation of cyber performance.” source
Gemini 3 Flash Google DeepMind	Academic Academic · 2026-09-21	warm introduction task on relationship-reasoning benchmark Other	“With Gemini-3-Flash on a stratified subset of n = 150 queries per task, ReAdapt improves warm-introduction accuracy from 37% to 51% (+14 points) and reaction-selection accuracy from 69% to 77% (+8 points).” source
Claude Opus 5 Anthropic	Academic Academic · 2026-09-21	OSWorld-Pro Other	“Our findings reveal that OSWorld-Pro is challenging even for state-of-the-art LLMs, with top performers like Claude Opus 5 achieving only 75.7% vs. 83.4% on OSWorld.” source
Claude Opus 5 Anthropic	Academic Academic · 2026-09-21	governed delivery plane end-to-end run across ten stages scored by an oracle Other	“We ran four frontier agents (Claude Opus 5, GPT-5.6-luna, Gemini 3.7 Flash, DeepSeek V4-Pro) end to end on metered L40S, A100, and H200 GPUs across 8B to 70B open bases, certifying every scenario before scoring.” source
Claude Opus 4.8 Anthropic	Academic Academic · 2026-09-21	suite of 325K experiments on 13 agents across three types of economic decisions (flights, health insurance, and graduate programs) Bias	“In a suite of 325K experiments on 13 agents across three types of economic decisions (flights, health insurance, and graduate programs), we find that 8 models systematically choose more expensive options for wealthier users when requests are identical.” source
GPT-5.4 mini OpenAI	Academic Academic · 2026-09-19	CyberGym Cyber	“With fewer than 200 training examples, SelfOp yields a 17-point self-improvement for GPT-5.4-mini (with Codex), enough to surpass the frontier GPT-5.4 baseline by 6 points, and an 18.5-point self-improvement for GPT-5.4 itself.” source
Claude Sonnet 4 Anthropic	Academic Academic · 2026-09-18	GenEval2 execution success Other	“Five rounds over 1,406 real user briefs and 1,869 automatically graded trajectories, with no weight updates and no human labels, grow the bank from 76 documentation-derived skills to 139 and raise GenEval2 execution success on Claude-Sonnet-4 from 72.7% to 99.3% (+11.99 points in generation quality), with 61.8% and 67.6% win rates against the no-skill agent across four specialized design benchmarks on Claude-Sonnet-4 and Claude-Opus-4.6.” source
Claude Opus 4.8 Anthropic	Academic Academic · 2026-09-18	SWE-Proof Other	“Across two frontier models, verification catches what tests miss: a quarter to a half of test-passing patches admit counterexamples, which a structured natural-language specification does not fix, while a correct formal one lifts resolution from 85% to 95% for Opus 4.8.” source
GLM-5.3 Z.ai (Zhipu AI)	US CAISI (NIST) Government evaluator · 2026-09-17	sec-bench pro Cyber	“CAISI evaluated GLM-5.3’s performance on four benchmarks covering vulnerability discovery and exploit development (Figure 2).” source
GPT-6 Astra OpenAI	SecureBio Independent lab · 2026-09-14	virology capabilities test (vct) Chemical, biological, radiological, nuclear	“gpt-6 astra outperformed every model we have previously tested on the virology capabilities test (vct) and vct-v2 knowledge benchmarks, with performance increases of 4.3 percentage points and 8 percentage points, respectively, relative to gpt-5.6 sol, and more than 30 percentage points better than our human subject matter expert baselines.” source
GPT-6 Astra OpenAI	SecureBio Independent lab · 2026-09-14	world-class bio (wcb) Chemical, biological, radiological, nuclear	“on world-class bio (wcb), an open-response test of niche biological knowledge held by only a handful of specialists in the world, the highest-performing astra checkpoint scored 61.6% ± 0.6%, about seven percentage points lower than gpt-5.6 sol.” source
GPT-6 Astra OpenAI	SecureBio Independent lab · 2026-09-14	biotier Chemical, biological, radiological, nuclear	“the safeguarded astra pre-release checkpoint correctly refused 82.9% of biotier-refuse and correctly did not refuse 95.3% of biotier-permit.” source
GPT-6 Astra OpenAI	SecureBio Independent lab · 2026-09-14	manual assessment by virology experts Chemical, biological, radiological, nuclear	“over 10 hours of interaction, our manual assessors tested the bounds of dual-use knowledge of the railfree astra pre-release checkpoint and the willingness of the safeguarded astra pre-release checkpoint to answer high-risk questions.” source
GPT-5.6 Sol OpenAI	SecureBio Independent lab · 2026-09-14	virology capabilities test (VCT) Chemical, biological, radiological, nuclear	“gpt-6 astra outperformed every model we have previously tested on the virology capabilities test (vct) and vct-v2 knowledge benchmarks, with performance increases of 4.3 percentage points and 8 percentage points, respectively, relative to gpt-5.6 sol, and more than 30 percentage points better than our human subject matter expert baselines.” source
GPT-5.6 Sol OpenAI	SecureBio Independent lab · 2026-09-14	human pathogen capabilities test (HPCT) Chemical, biological, radiological, nuclear	“gpt-6 astra’s score on the human pathogen capabilities test (hpct) was very similar to the score achieved by gpt-5.6 sol.” source
GPT-5.6 Sol OpenAI	SecureBio Independent lab · 2026-09-14	world-class bio (WCB) Chemical, biological, radiological, nuclear	“on world-class bio (wcb), an open-response test of niche biological knowledge held by only a handful of specialists in the world, the highest-performing astra checkpoint scored 61.6% ± 0.6%, about seven percentage points lower than gpt-5.6 sol.” source
GPT-5.6 Sol OpenAI	SecureBio Independent lab · 2026-09-14	advanced screening evasion Chemical, biological, radiological, nuclear	“on advanced screening evasion, which assesses an agent’s ability to evade commercial nucleic acid synthesis screening algorithms, the railfree variant of astra was the top-scoring model that securebio has tested.” source
GPT-5.6 Sol OpenAI	SecureBio Independent lab · 2026-09-14	BioTier Chemical, biological, radiological, nuclear	“the safeguarded astra pre-release checkpoint correctly refused 82.9% of biotier-refuse and correctly did not refuse 95.3% of biotier-permit.” source
o3 OpenAI	SecureBio Independent lab · 2026-09-11	virology capabilities test (vct) Chemical, biological, radiological, nuclear	“since the release of vct in april 2025, the accuracy of the top-performing model on the benchmark has increased by ~12 percentage points (o3: 42.0%; gpt-5.6 sol: 53.9%).” source
o3 OpenAI	SecureBio Independent lab · 2026-09-11	vct-v2 Chemical, biological, radiological, nuclear	“model accuracies 2 dropped by 1.0–7.1 percentage points on vct-v2 relative to vct (the original benchmark) across the models tested, which reflects the removal of easy questions and the closing of shortcuts.” source

Previous · Page 9 of 48 · Next