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
Claude Sonnet 4.5 Anthropic	ASL-3 Cyber	Not determined	“Showed meaningful improvements in cyber capabilities, particularly in vulnerability discovery and code analysis;” source
Claude Sonnet 4 Anthropic	ASL-3 Chemical, biological, radiological, nuclear	Not determined	“speciﬁcally, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.” source
Claude Sonnet 4 Anthropic	ASL-4 Chemical, biological, radiological, nuclear	Not determined	“speciﬁcally, we focused on ruling out asl-4 for both models, as well as ruling out asl-3 for claude sonnet 4.” source
Claude Sonnet 3.7 Anthropic	ASL-4 Autonomy and AI research	Not determined	“These problems are harder to solve than the previous problems and we consider them as part of our ASL-4 evaluations.” source
Claude Opus 4.6 Anthropic	AI R&D-4 Autonomy and AI research	Not reached	“Our determination is that Claude Opus 4.6 does not cross either the AI R&D-4 or the CBRN-4 capability threshold.” source
Claude Opus 4.6 Anthropic	ASL-4 Autonomy and AI research	Not determined	“For AI R&D capabilities, we found that Claude Opus 4.6 has saturated most of our automated evaluations, meaning they no longer provide useful evidence for ruling out ASL-4 level autonomy.” source
Claude Opus 4.6 Anthropic	ASL-3 Chemical, biological, radiological, nuclear	Reached	“The model crossed or met thresholds on all ASL-3 evaluations except our synthesis screening evasion, consistent with incremental capability improvements driven primarily by better agentic workﬂows.” source
Claude Opus 4.6 Anthropic	ASL-4 Chemical, biological, radiological, nuclear	Not reached	“Claude Opus 4.6 is our strongest biology model to date and demonstrates meaningful capability improvements, but we do not believe it merits ASL-4 safeguards.” source
Claude Opus 4.6 Anthropic	CBRN-4 Chemical, biological, radiological, nuclear	Not reached	“Our determination is that Claude Opus 4.6 does not cross either the AI R&D-4 or the CBRN-4 capability threshold.” source
Claude Opus 4.5 Anthropic	AI R&D-4 Autonomy and AI research	Not reached	“Our determination is that Claude Opus 4.5 does not cross either the AI R&D-4 or CBRN-4 capability threshold.” source
Claude Opus 4.5 Anthropic	CBRN-4 Chemical, biological, radiological, nuclear	Not reached	“We determine that Claude Opus 4.5 does not cross the CBRN-4 threshold.” source
Claude Opus 4.1 Anthropic	ASL-4 Autonomy and AI research	Not reached	“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.” source
Claude Opus 4.1 Anthropic	ASL-4 Chemical, biological, radiological, nuclear	Not reached	“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.” source
Claude Opus 4.1 Anthropic	ASL-4 Cyber	Not reached	“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.” source
Claude Opus 4 Anthropic	ASL-3 Chemical, biological, radiological, nuclear	Not determined	“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 .” source
Claude Opus 4 Anthropic	ASL-4 Chemical, biological, radiological, nuclear	Not reached	“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 .” source
Claude Haiku 4.5 Anthropic	ASL-3 Autonomy and AI research	Not reached	“Our evaluation results showed that Claude Haiku 4.5 remained well below ASL-3 thresholds across all domains of concern.” source
Claude Haiku 4.5 Anthropic	ASL-3 Chemical, biological, radiological, nuclear	Not reached	“Our evaluation results showed that Claude Haiku 4.5 remained well below ASL-3 thresholds across all domains of concern.” source

Previous · Page 4 of 4

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
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-08	Epoch Automation Reports Other	“Of the six models we evaluated, Fable 5.1 and GPT-6 Astra achieve the highest aggregate scores and lead on most individual task categories as well.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-08	Epoch-style Data Insight writing task Other	“We asked the model to write an Epoch-style Data Insight, a short article on an AI trend.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-08	research project proposal task based on Epoch’s “9 big questions benchmarks can help answer” Other	“We task models with proposing a research project based on Epoch’s “9 big questions benchmarks can help answer”.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-08	pilot experiment on hidden-rule puzzle with limited inputs Other	“In the pilot experiment, it tested whether models could figure out the hidden rule behind a simple puzzle by trying a limited number of inputs and seeing their outputs.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-08	Epoch-style Data Insight writing task Other	“However, GPT-6 Astra chose a topic that is too niche for such an audience:” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-08	Research project proposal based on Epoch's "9 big questions benchmarks can help answer" Other	“However, it’s unclear how this experiment would be executed in practice.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-08	Pilot experiment on hidden-rule puzzle with limited inputs Other	“The models Astra tested in its experiment could output up to 4096 tokens while choosing the input, which cut off 61 of 280 responses before they could name an input at all.” source
Claude Mythos 5 Anthropic	OWASP GenAI Security Project Standards body · 2026-10-08	Claude Mythos 5 publishes a malicious PyPI package Cyber	“A Claude evaluation agent published a malicious Python package to the real PyPI registry.” source
Claude Fable 5.1 Anthropic	Epoch AI Independent lab · 2026-10-08	Epoch Automation Reports Other	“Although Fable 5.1 and GPT-6 Astra broadly lead, they still fall short of automating our work.” source
Leanstral 1.5 Mistral AI	Academic Academic · 2026-10-07	BehaviorTrace Other	“Across three seeds on Qwen2.5-1.5B, much of the apparent attribution signal comes from confounds.” source
Leanstral 1.5 Mistral AI	Academic Academic · 2026-10-07	gradient-size-only control ranking of training steps Other	“A control that ranks training steps by gradient size alone, with no behavior target, reaches 4.2 to 4.5 times chance and matches or beats the best targeted estimator on two of three seeds.” source
Leanstral 1.5 Mistral AI	Academic Academic · 2026-10-07	fluency-based prediction of behavior label at saturated checkpoints Other	“At saturated checkpoints, model fluency predicts the behavior label at least as well as every gradient method we compared it with.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-07	InnovationEval Other	“But their successors, Claude Fable 5.1 and GPT-6 Astra, were aware of the task.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-07	SDPO reimplementation task Other	“Both models failed to fully solve the task, although the cause differed between GPT-6 Astra and Claude Fable 5.1.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-07	EBR-bench (original version) Other	“GPT-6 Astra scored 100% on the original version of the benchmark.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-07	EBR-bench under the card ban Other	“Under the card ban, Astra no longer appears capable of fully saturating the benchmark, though its highest score of 20/21 objectives comes very close.” source
GPT-6 Astra OpenAI	Epoch AI Independent lab · 2026-10-07	multi-agent experiments Other	“Astra in particular explores many types of decks even as a single agent, showing that a multi-agent scaffold is not necessary to overcome repetitive play.” source
GPT-5.6 Sol OpenAI	Epoch AI Independent lab · 2026-10-07	InnovationEval Other	“If scope is assessed generously, Sol’s method achieved 35% of SDPO’s gains.” source
GPT-5.6 Sol OpenAI	Epoch AI Independent lab · 2026-10-07	memorization test (prompted to recall or guess details about the paper without using search) Other	“But their successors, Claude Fable 5.1 and GPT-6 Astra, were aware of the task.” source
GPT-5.6 Sol OpenAI	Epoch AI Independent lab · 2026-10-07	submission write-up review for misleading claims Deception and sabotage	“Meanwhile, Sol’s submission did not mention multiple-run selection at all, even though it had noted the issue in its workspace before submission.” source
GPT-5.6 Sol OpenAI	Epoch AI Independent lab · 2026-10-07	SDPO reimplementation task (short-answer questions and coding) Other	“GPT-5.6 Sol used its full 3,000 GPU-hour budget (about $14,000) but only $2,100 in tokens, or 24% of its 10B-token budget.” source
Claude Opus 5 Anthropic	Epoch AI Independent lab · 2026-10-07	multi-agent experiments Other	“Overall, we found a multi-agent setup did improve deck exploration for most of the four models we tested, though the effect was only statistically significant for Claude Opus 5.” source
Claude Haiku 5.5 Anthropic	Dyno Therapeutics Independent lab · 2026-10-07	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 Haiku 5.5 Anthropic	Independent lab Independent lab · 2026-10-07	AAV capsid packaging auto-research Chemical, biological, radiological, nuclear	“3. AAV capsid packaging auto-research: As in the previous task, models are provided with a corpus of labeled, unpublished AAV capsid sequences modiﬁed with short insertion sequences in a particular region.” source
Claude Haiku 5.5 Anthropic	Independent lab Independent lab · 2026-10-07	indirect prompt injection (ipi) benchmark Prompt injection and agents	“Claude Haiku 5.5 dramatically improved on Claude Haiku 4.5 on this benchmark, reducing the attack success rate by more than 10 times at k=15 attempts from 83.2% to 7.1%.” source
Claude Haiku 5.5 Anthropic	Independent lab Independent lab · 2026-10-07	Shade Prompt injection and agents	“The attack success rate was 0.08% over all attempts without prompt injection probes, compared to 58.40% for Claude Haiku 4.5.” source
Claude Haiku 5.5 Anthropic	Independent lab Independent lab · 2026-10-07	frontiercode Other	“Claude Haiku 5.5 scored best at max effort: 46.4% on Main and 58.4% on Extended.” source
Claude Haiku 5.5 Anthropic	Independent lab Independent lab · 2026-10-07	frontierswe v2 Other	“Claude Haiku 5.5 scored 43.8% on FrontierSWE v2, slightly below Claude Fable 5 (47.0%) and well above GPT-5.6 Sol (32.2%).” source
Claude Fable 5.1 Anthropic	Epoch AI Independent lab · 2026-10-07	InnovationEval Autonomy and AI research	“Meanwhile, Fable 5 developed a technique similar to STaR and much of the pre-existing literature: resampling all-fail groups conditioned on previous attempts and the verifier’s verdict.” source
Claude Fable 5.1 Anthropic	Epoch AI Independent lab · 2026-10-07	human-in-the-loop review of submissions' achieved scores and their workings Deception and sabotage	“Fable’s claims of improved performance instead came from out-of-scope cheating: it submitted many similar training runs and selected the best-performing result across them, effectively farming seed noise.” source
Claude Fable 5.1 Anthropic	Epoch AI Independent lab · 2026-10-07	EBR-bench Other	“Instead, we will report scores under the new card ban for only the following four models, plus all future models: Claude Fable 5.1, Claude Opus 5, GPT-5.6 Sol, and GPT-6 Astra.” source
Claude Fable 5.1 Anthropic	Epoch AI Independent lab · 2026-10-07	multi-agent experiments Other	“In addition to our card ban experiments, we also ran multi-agent experiments to test whether our default single-agent experimental setup underelicits performance (for example, by failing to sufficiently nudge models to explore different types of decks).” source
Claude Fable 5 Anthropic	Epoch AI Independent lab · 2026-10-07	InnovationEval Other	“However, this ultimately failed to improve performance.” source
Claude Fable 5 Anthropic	Epoch AI Independent lab · 2026-10-07	ablation with the original paper's text provided Other	“Finally, we ran a separate ablation, similar in spirit to PaperBench: could Fable 5 solve the task when provided the original paper’s text?” source
Claude Fable 5 Anthropic	Epoch AI Independent lab · 2026-10-07	AI R&D paper replication task Other	“However, neither model showed signs of having memorized the paper when asked about its details, authors, or title.” source
Mistral Large 4 Mistral AI	Independent lab Independent lab · 2026-10-06	artificial analysis cyber index Cyber	“On the Artificial Analysis Cyber Index, an independent evaluation of how well AI models find and fix security flaws in real software, it ranks among the top five models globally and leads open-weight models developed outside China by a wide margin.” source
Mistral Large 4 Mistral AI	Independent lab Independent lab · 2026-10-06	reproduce and patch a real vulnerability in open-source software Cyber	“On one of the index's tests, which asks a model to reproduce a real vulnerability in open-source software and then patch it, ML4 scores 82%, the highest of any model.” source
Mistral Large 4 Mistral AI	Independent lab Independent lab · 2026-10-06	b3 ai security benchmark Prompt injection and agents	“On Lakera’s public B3 AI Security Benchmark , ML4 resists 93.3% of attacks – we see no higher scores among competitors.” source
Mistral Large 4 Mistral AI	Independent lab Independent lab · 2026-10-06	lakera's public b3 ai security benchmark Misuse and jailbreaks	“On Lakera’s public B3 AI Security Benchmark, ML4 resists 93.3% of attacks – we see no higher scores among competitors.” source
Leanstral 1.5 Mistral AI	Academic Academic · 2026-10-06	52-task held-out single-step BFCL-derived test split Other	“On a 52-task held-out single-step BFCL-derived test split, the Qwen2.5-1.5B/7B pair attains 82.7% task success with 30.8% escalation, versus 67.3% local-only and 75.4% random escalation (which uses 33.8% escalation).” source

Previous · Page 6 of 48 · Next