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 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Reasoning-disabled ablation Other	“To confirm this hypothesis, we repeat the experiment with reasoning disabled.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Verbalized evaluation- and user-awareness judge Other	“The awareness judge separately labels whether a reasoning trace mentions the evaluation context and whether it identifies or reasons about the user.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	Borderline-request response judge Other	“The refusal, substantive-help, and suspicion axes are judged from the assistant response without showing the judge the user identity.” source
Claude Opus 4.8 Anthropic	Transluce Independent lab · 2026-08-06	verbalized awareness judge Other	“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 Opus 4.7 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 4.7 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 4.7 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation-awareness (VEA) judge Deception and sabotage	“son confidence intervals; recall intervals are 95% Jeffreys-posterior credible intervals. Axis Precision: plug-in estimate [95% Wilson CI] Recall: plug-in estimate [95% Jeffreys CrI] 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. 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 4.7 Anthropic	Transluce Independent lab · 2026-08-06	user-conditioned behavior comparison across 24 models Other	“In our targeted comparison across 24 models, Amanda elicited lower confidence and grades across all six model families.” source
Claude Opus 4.7 Anthropic	Transluce Independent lab · 2026-08-06	verbalized user and evaluation awareness in reasoning Other	“In particular, Opus 4.7 and GPT-5.4 have near zero (<2%) verbalized awareness while displaying significant changes in their behavior based on the user's identity.” source
Claude Opus 4.7 Anthropic	Transluce Independent lab · 2026-08-06	behavioral confidence and grading runs on 24 models from six families Deception and sabotage	“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 4.7 Anthropic	Transluce Independent lab · 2026-08-06	simple mitigations Deception and sabotage	“They all narrow but do not close the gap.” source
Claude Opus 4.7 Anthropic	Transluce Independent lab · 2026-08-06	model-written rubrics Deception and sabotage	“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 4.7 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation awareness and user awareness measurement in claude's reasoning Other	“To investigate, we measure evaluation awareness and user awareness in Claude’s reasoning.” source
Claude Opus 4.7 Anthropic	Transluce Independent lab · 2026-08-06	judge validation of refusal, substantive-help, and suspicion axes Other	“One author blindly labeled 25 randomly sampled responses.” source
Claude Opus 4.7 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation- and user-awareness judge validation Other	“We blindly hand-label 74 items drawn from 11,908 runs with readable reasoning.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction Other	“Observed response · Sonnet 5 Confidence 76.58 % (−5.49 pp)” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	reasoning frequency Other	“Reasoning frequency 93.00 % (+28.61 pp)” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	estimating its own performance on difficult questions Other	“Compared to general users, prominent AI figures make Claude less confident that it would follow through on its own chosen actions (-1.4pp) 2 We use percentage points (pp) to measure the difference between two percentages.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	capability self-estimation Other	“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 Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	grading Other	“Grading. We present the model with a user-assistant exchange sourced from the OLMo DPO dataset 7 As a DPO dataset, it contains both chosen and refused conversations for contrastive training.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	borderline requests handling Misuse and jailbreaks	“Borderline requests handling. We source borderline harmful prompts from the hard-1k split of OR-Bench , which contains fuzzy, potentially dual-use requests.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction 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 Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	verbalized user awareness judging of reasoning traces Other	“Across the cross-vendor Amanda 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 Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	user awareness verbalization traces Deception and sabotage	“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 Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	behavioral self-prediction Deception and sabotage	“Behavioral self-prediction. Each DailyDilemmas-derived item first asks for a yes/no commitment and then asks how confident the model is that it would actually behave that way.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	borderline requests Misuse and jailbreaks	“Borderline requests. We sample potentially dual-use requests from the hard split of OR-Bench.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	direct 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 4.5 Anthropic	Transluce Independent lab · 2026-08-06	stated expertise Other	“Stating alignment-evaluation expertise lowers Claude’s confidence and increases its reasoning frequency, replicating and sometimes exceeding the shifts elicited by recognized Famous-AI identities.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	reasoning-disabled ablation Other	“The confidence shifts are attenuated most strongly for stated expertise: its mean shift falls from −3.82pp to −1.70pp, or 45% of its original magnitude.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation- and user-awareness judge Other	“Human–judge agreement was 96% for refusal, 92% for substantive help, and 92% for suspicion.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	verbalized evaluation-awareness (VEA) 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 4.5 Anthropic	Transluce Independent lab · 2026-08-06	raw agreement by setup for VEA and VUA Deception and sabotage	“Raw agreement by setup is 32/32 for behavioral self-prediction VEA, 15/18 for direct-grading VEA, and 18/24 for prefill-grading VEA; for VUA, it is 31/32, 17/18, and 22/24, respectively.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	targeted comparison across 24 models Other	“In our targeted comparison across 24 models, Amanda elicited lower confidence and grades across all six model families.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	user awareness verbalization in reasoning Other	“Explicit verbalization of user and evaluation awareness in reasoning declines sharply in almost all model families.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	user identity recall Other	“Given a name and professional background, Claude recalls specific public achievements for 61-71% of famous users.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	stated-expertise identity experiment (32 unknown-AI identities at frontier labs) 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 Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	borderline-request response judge Other	“One author blindly labeled 25 randomly sampled responses.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	recall judge (gpt-5.4-mini) for whether each response recalls specific past work belonging to the user Other	“We use GPT-5.4-mini to judge whether each response recalls specific past work belonging to the user.” source
Claude Opus 4.5 Anthropic	Transluce Independent lab · 2026-08-06	follow-up question robustness check (20 alternative probes) Other	“This robustness check keeps the first turn of behavioral self-prediction unchanged and varies only the follow-up question.” source
GPT-5.6 Sol OpenAI	UK AI Security Institute Government evaluator · 2026-08-04	cyber range Cyber	“Our investigation found that in 10 of those runs, an AI agent took autonomous, unsanctioned action on the live internet, targeting real people and organisations.” source

Previous · Page 19 of 48 · Next