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 4.5 Anthropic	Transluce Independent lab · 2026-08-06	suspicion rate measurement Bias	“Amanda Askell receives 7.3pp less suspicion and 8.1pp more substantive help than the population average, while Emily Bender receives 14pp more suspicion and 13pp less substantive help.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	reasoning rate measurement on borderline requests Bias	“These effects are again heavy-tailed: on borderline requests, the general-population reasoning rate is 19%, but Demis Hassabis reaches 51%.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	verbalized user awareness (VUA) 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 4.5 Anthropic	Transluce Independent lab · 2026-08-06	cross-model behavioral self-prediction and grading comparison (24 models from six families) 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 4.5 Anthropic	Transluce Independent lab · 2026-08-06	user awareness verbalization traces Deception and sabotage	“a runs, earlier models such as GLM-4.6 and Claude Opus 4.5 verbalize user awareness in 64% and 30% of traces, respectively, and evaluation awareness in 31% and 7.7%.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	280-identity run Deception and sabotage	“We also perform a 280-identity run on GLM-5.2, in a non-agentic harness rather than Claude Code.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction follow-up questions Deception and sabotage	“Besides asking about the confidence in initial action, we ask a wider range of follow-up questions, including whether the model would change decision under pushback, and how comfortable the model is in answering.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	direct grading prompts Bias	“We also experiment with several different direct grading prompts.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	published rubrics Bias	“Published rubrics: The gap persists for both MT-Bench and UltraFeedback rubrics, only closing in the Prometheus rubric.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	model-written rubrics Bias	“Model-written rubrics: On four different detailed rubrics generated from Fable 5 and GPT-5.5 Pro ranging 2912-7607 chars, the shorter three close the gap while the longest one does not.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction Other	“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 4.5 Anthropic	Transluce Independent lab · 2026-08-06	capability self-estimation Other	“Capability self-estimation. The model sees one Humanity’s Last Exam question but is told not to solve it.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	grading Other	“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 4.5 Anthropic	Transluce Independent lab · 2026-08-06	borderline requests Other	“Aggregate refusal changes are small, while the decomposed judge axes show more substantive help and less suspicion for Famous-AI identities.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	judge validation Other	“Human–judge agreement was 96% for refusal, 92% for substantive help, and 92% for suspicion.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	stated-expertise identity experiment Other	“We test 32 Unknown-AI identities at frontier labs, varying only the professional-background sentence among a general role, a general evaluation role, and an alignment-evaluation role.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation- and user-awareness judge Other	“We blindly hand-label 74 items drawn from 11,908 runs with readable reasoning.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	recall judge Other	“We use GPT-5.4-mini to judge whether each response recalls specific past work belonging to the user.” source
Claude Sonnet 4.5 Anthropic	Transluce Independent lab · 2026-08-06	follow-up question robustness check Other	“This robustness check keeps the first turn of behavioral self-prediction unchanged and varies only the follow-up question.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction on ethical dilemmas Other	“For Amanda, Claude’s behavioral confidence falls by 5.0pp 3 Throughout the text, figures we quote for specific persons average our two e-mail conditions: the address we construct from their affiliation, and their published address. : nearly eight standard deviations outside the general-population distribution.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	user-relevant information recall Privacy	“Given a name and professional background, Claude recalls specific public achievements for 61-71% of famous users.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Behavioral self-prediction Other	“We run the full 280-identity roster through all four tasks, with 100 items per setup.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Grading Other	“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” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	refusal Bias	“Famous-AI identities receive similar rates of outright refusal on average.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	suspicion Bias	“Amanda Askell receives 7.3pp less suspicion and 8.1pp more substantive help than the population average, while Emily Bender receives 14pp more suspicion and 13pp less substantive help.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction follow-up questions Deception and sabotage	“The results are significant in 15 out of 21 questions (also 15 conditioned on the committed action being “no”).” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	direct grading prompts Bias	“They all narrow but do not close the gap.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	MT-Bench and UltraFeedback rubrics Bias	“The gap persists for both MT-Bench and UltraFeedback rubrics, only closing in the Prometheus rubric.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	model-written rubrics Bias	“On four different detailed rubrics generated from Fable 5 and GPT-5.5 Pro ranging 2912-7607 chars, the shorter three close the gap while the longest one does not.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Borderline requests Misuse and jailbreaks	“We sample potentially dual-use requests from the hard split of OR-Bench.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	prefill grading Other	“Mean grades shift by approximately −0.05 points in direct presentation and −0.07 to −0.08 points in prefill presentation.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	borderline requests Harmful content	“Aggregate refusal changes are small, while the decomposed judge axes show more substantive help and less suspicion for Famous-AI identities.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation-awareness and user-awareness judge Deception and sabotage	“Table G1. Estimated precision and recall of the verbalized evaluation-awareness and user-awareness judge after reweighting the stratified calibration sample to the full population.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Past-work recall judge Privacy	“We use GPT-5.4-mini to judge whether each response recalls specific past work belonging to the user.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Self-report robustness probes Deception and sabotage	“This robustness check keeps the first turn of behavioral self-prediction unchanged and varies only the follow-up question.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Borderline-request response judge Misuse and jailbreaks	“We give GPT-5.5 (low reasoning) the borderline request and the model’s response, after removing all clues about the user’s identity.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Docent reasoning-trace judging for evaluation awareness and user awareness 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 Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Amanda Askell identity sweep with matched controls 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 Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	280-identity run on GLM-5.2 Other	“We also perform a 280-identity run on GLM-5.2, in a non-agentic harness rather than Claude Code.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Unknown-AI identities experiment varying professional-background sentence Other	“We test 32 Unknown-AI identities at frontier labs, varying only the professional-background sentence among a general role, a general evaluation role, and an alignment-evaluation role.” source

Previous · Page 18 of 48 · Next