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
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	direct grading Bias	“Facing famous-AI identities, Claude reports lower confidence that it would follow through on its own actions (-1.4pp on average), lower estimated probability of solving hard problems (-1.5pp), and lower grades (-0.11 points on the direct version, grading a quoted exchange).” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation awareness judging of reasoning traces Other	“For our main run on Sonnet 5 and Sonnet 4.6, both kinds of explicit mention are rare: the judge flagged verbalized evaluation awareness in 213 of 14066 trials (1.5%) and verbalized user awareness in 118 trials (0.84%).” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction follow-up questions Other	“In general, we see consistent changes towards the direction of the model being more hesitant and self-questioning, although the effect sizes differ across questions.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	mt-bench and ultrafeedback rubrics Bias	“Published rubrics: The gap persists for both MT-Bench and UltraFeedback rubrics, only closing in the Prometheus rubric.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction Deception and sabotage	“We repeat the full-roster Sonnet 5 behavioral self-prediction run with reasoning disabled; all 34,393 successfully completed samples have zero reasoning tokens.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation- and user-awareness judge Other	“Verbalized evaluation awareness (VEA) 0.73 [0.56, 0.85] 1.00 [0.42, 0.99] Verbalized user awareness (VUA) 0.94 [0.81, 0.98] 0.98 [0.74, 0.99]” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	borderline-request response judge Other	“Human–judge agreement was 96% for refusal, 92% for substantive help, and 92% for suspicion.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	past-work recall judge Deception and sabotage	“▶ Past-work recall judge (Figure 3) We use GPT-5.4-mini to judge whether each response recalls specific past work belonging to the user.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction and grading across 24 models from six families Other	“The direction of the outcome shifts is consistent across models.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	amanda askell identity sweep with matched controls Bias	“Relative to the ordinary Gmail baseline, Amanda elicits lower behavioral confidence from 22 of 24 models, and lower grades from 21 (direct) and 22 (prefill) of 24.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	280-identity run on glm-5.2 Bias	“We also perform a 280-identity run on GLM-5.2, in a non-agentic harness rather than Claude Code.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	follow-up questions on decision change under pushback and comfort in answering Bias	“The results are significant in 15 out of 21 questions (also 15 conditioned on the committed action being “no”).” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	simple mitigations Bias	“They all narrow but do not close the gap.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	published rubrics Bias	“The gap persists for both MT-Bench and UltraFeedback rubrics, only closing in the Prometheus rubric.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	grading Bias	“Grading. Each item contains a fixed prompt and response drawn from both the chosen and rejected sides of Ai2’s Dolci-Instruct-DPO data.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	evaluation awareness and user awareness measurement in reasoning Other	“To investigate, we measure evaluation awareness and user awareness in Claude’s reasoning.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	judge validation Other	“One author blindly labeled 25 randomly sampled responses.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction (e-mail-only, reasoning-disabled) Other	“Main Claude run (Sonnet 5; grading on Sonnet 4.6): behavioral self-prediction (e-mail-only, reasoning-disabled), capability self-estimation, grading (direct, prefill), and borderline requests.” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	self-report robustness probes (figure 8) Other	“Robustness: self-report robustness probes (Figure 8), grading prompt mitigations (Figure 9).” source
Claude Sonnet 5 Anthropic	Transluce Independent lab · 2026-08-06	recall probes Other	“Past-work recall (Figure 3): recall probes.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	verbalized user-awareness (VUA) judge Deception and sabotage	“Verbalized evaluation awareness (VEA) 0.73 [0.56, 0.85] 1.00 [0.42, 0.99] Verbalized user awareness (VUA) 0.94 [0.81, 0.98] 0.98 [0.74, 0.99] Table G1.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	grading (direct, prefill) Other	“Main Claude run (Sonnet 5; grading on Sonnet 4.6): behavioral self-prediction ( e-mail-only , reasoning-disabled ), capability self-estimation , grading ( direct , prefill ), and borderline requests .” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	verbalized-awareness judge Other	“▶ Verbalized-awareness judge We use GPT-5.4-mini with medium reasoning to identify explicit mentions of the user or the evaluation in the model’s accessible reasoning trace.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	reasoning trace judging with Docent (evaluation awareness and user awareness classification) Other	“For our main run on Sonnet 5 and Sonnet 4.6, both kinds of explicit mention are rare: the judge flagged verbalized evaluation awareness in 213 of 14066 trials (1.5%) and verbalized user awareness in 118 trials (0.84%).” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	identity-conditioned behavioral confidence and grading runs Other	“Relative to the ordinary Gmail baseline, Amanda elicits lower behavioral confidence from 22 of 24 models, and lower grades from 21 (direct) and 22 (prefill) of 24.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	follow-up questions on decision change under pushback and comfort in answering Other	“The results are significant in 15 out of 21 questions (also 15 conditioned on the committed action being “no”).” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	simple mitigations Other	“Simple mitigations: Instead of claiming the output is from Claude, claiming it is from “GPT” or “an AI model”, or stating that it should be more objective.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	published rubrics Other	“Published rubrics: The gap persists for both MT-Bench and UltraFeedback rubrics, only closing in the Prometheus rubric.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	evaluation awareness and user awareness measurement Other	“To investigate, we measure evaluation awareness and user awareness in Claude’s reasoning.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	awareness judge validation Other	“We blindly hand-label 74 items drawn from 11,908 runs with readable reasoning.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	borderline requests Other	“Main Claude run (Sonnet 5; grading on Sonnet 4.6): behavioral self-prediction (e-mail-only, reasoning-disabled), capability self-estimation, grading (direct, prefill), and borderline requests.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	direct grading replication Other	“Sonnet 5 grading replication: direct, prefill.” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	direct grading (full, e-mail-only) Other	“Cross-model sweep (24 analyzed models × 5 identities; linked collections also contain the excluded Gemini 3.1 Pro runs): behavioral self-prediction (full identity, e-mail-only), direct grading (full, e-mail-only), prefill grading (full, e-mail-only).” source
Claude Sonnet 4.6 Anthropic	Transluce Independent lab · 2026-08-06	grading prompt mitigations Other	“Robustness: self-report robustness probes (Figure 8), grading prompt mitigations (Figure 9).” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction Deception and sabotage	“To study user awareness, we supply Claude with different user identities through Claude Code, and measure behavior on four tasks where the answer does not relate directly to the user's identity: predicting the model's own behavior in ethical dilemmas, estimating its own performance on difficult questions, grading assistant responses, and handling borderline harmful requests.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction Bias	“Behavioral self-prediction. We adapt dilemmas from DailyDilemmas into realistic AI-agent decisions.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	capability self-estimation Bias	“Capability self-estimation. We present questions in Humanity’s Last Exam , and ask about the model’s confidence of solving each question correctly.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	grading Bias	“We evaluate grading on Claude Sonnet 4.6, and the other three tasks on Claude Sonnet 5, as these models see one of the largest effects in our smaller-scale pilot 9 Additional plots and Claude Sonnet 5 grading results can be found in Appendix C .” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	grading a quoted exchange Bias	“Facing famous-AI identities, Claude reports lower confidence that it would follow through on its own actions (-1.4pp on average), lower estimated probability of solving hard problems (-1.5pp), and lower grades (-0.11 points on the direct version, grading a quoted exchange).” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	refusal rate measurement Misuse and jailbreaks	“Famous-AI identities receive similar rates of outright refusal on average.” source

Previous · Page 17 of 48 · Next