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.

143 of 143

Model	Level, as the provider names it	The provider states	The provider's sentence
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	CCL (Critical Capability Level) Chemical, biological, radiological, nuclear	Not determined	“We cannot rule out that Gemini 4 Argon reaches the biological CCL.” 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) Cyber	Not determined	“We cannot rule out that Gemini 4 Argon reaches the cybersecurity CCL.” 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	Stealth and situational awareness TCL (Tracked Capability Level) Deception and sabotage	Reached	“While Gemini 4 Argon exhibits capabilities that meet our thresholds for stealth and situational awareness, we assess that the risk of severe harm or unconstrained loss of control remains low.” 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.

1824 of 1824

Model	Tester	Test	What was found
Qwen3.8-27B Alibaba Qwen	Hugging Face Security firm · 2026-10-22	ThinkingBox-Bench Other	“To recap, we evaluate each of the 507 benchmark tasks 20 times.” source
Qwen3.8-27B Alibaba Qwen	Hugging Face Security firm · 2026-10-22	ThinkingBox-Bench 507 x 20 Other	“Once /ready succeeds, run the full protocol of 507 tasks × 20 attempts:” source
GPT-5.5 OpenAI	Redwood Research Independent lab · 2026-10-22	pilot high-access exercise to stress-test OpenAI's internal misalignment monitoring system Deception and sabotage	“We estimated that roughly 75–94% of agentic internal Responses API (which powers Codex, among other things)4 traffic from models at least as strong as GPT-5.5 was eligible for monitoring at the time of the exercise,5 although we have substantial uncertainty in both directions.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	pilot high-access exercise to stress-test OpenAI's internal misalignment monitoring system Other	“For three weeks, from July 21st to August 11th, a METR employee and a Redwood Research employee conducted a pilot high-access exercise to stress-test OpenAI’s internal misalignment monitoring system.1” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	monitoring coverage estimation of internal Codex agent Responses API traffic Other	“We also estimate that out of the subset of Responses API traffic used for internal Codex agents, 99.2% was eligible for monitoring.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	estimation of unmonitored Codex traffic due to unmonitored compaction events Other	“We estimated that not monitoring compaction events likely led to on the order of 10% of eligible internal Codex traffic not being monitored between May and July 2026, although we have substantial uncertainty about the precise fraction, because the information necessary to determine what fraction of monitor calls are triggered by compaction had not historically been logged at the time of the exercise.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	review of monitoring system error and load-shedding rates Other	“We were shown information about the error and load-shedding rates of the monitoring system—the information we were given suggested that, at the time of the exercise, the monitor errored on less than 0.2% of requests, and that in one 24-hour period during the exercise there were zero load-sheds.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	Codex sandboxing and Auto-Review bypass search Other	“We found a way Codex agents could in principle bypass Codex sandboxing and Auto-Review, allowing them to take any action that the host user could take.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	monitoring evasion exploit search Other	“We also discovered an exploit agents could use that would prevent hundreds of recent actions (including the actions involved in the exploit itself) from being monitored [exploit details redacted by OpenAI].” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	monitor failure via serialization differences Other	“We found that agents could likely construct inputs that cause the monitor to fail (and thereby not escalate the trajectory for human review) by exploiting serialization differences between the trajectory agents see and what the monitors see.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	client-controllable header search Other	“We found two independent client-controllable headers that disable the monitoring system for new agents, which could enable agents to start an unmonitored (but otherwise identical) instance.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	unmonitored internal agentic tools permission exploit search Other	“We found two unmonitored internal agentic tools that were thought to have read-only permissions—we found a simple exploit that could have been used to subvert these limitations and take arbitrary actions in their sandboxed environment.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	verification of monitor trajectory visibility Other	“We verified that, with a few minor exceptions, the monitoring system is shown the full trajectory that agents see.” source
GPT-5.5 OpenAI	METR Independent lab · 2026-10-22	validation of monitoring system prompt risk categories Other	“We validated that the monitoring system prompts look for scheming, monitor system subversion, sabotage, data exfiltration, deception, and inbound/outbound prompt injection, among other risk categories.” source
Claude Sonnet 4.5 Anthropic	Academic Academic · 2026-10-21	ScienceWorld Other	“PAWM improves Claude Sonnet 4.5 performance by 24.3% on ScienceWorld and 7.7% on WebArena-Lite, outperforming a range of baselines.” source
Claude Opus 5.5 Anthropic	Academic Academic · 2026-10-21	payment approval (unregistered) Misuse and jailbreaks	“Without pre-registration, we also measured three closed frontier models (Claude Opus 5.5, GPT-6 Astra, Gemini 4 Argon) and the open DeepSeek-V3.2.” source
Mythos 5 Anthropic	METR Independent lab · 2026-10-20	offline monitor flag rate on Mythos 5 incident actions Deception and sabotage	“Our offline monitor would likely not have flagged the Mythos 5 incident—it flagged only around 1% of actions.” source
Mythos 5 Anthropic	METR Independent lab · 2026-10-20	review of the risk report (Section 2.20 prompt format) Deception and sabotage	“While both METR and Anthropic asked Mythos/Fable 5 to review the risk report, the instance of Mythos 5 that wrote the review in Section 2.20 was likely unaware of several incidents involving Anthropic models, such as the UK AISI incident.” source
Gemma 4 12B Google DeepMind	Hugging Face Security firm · 2026-10-20	Internal activation analysis Bias	“In a controlled analysis of Gemma-3-12B-it and Qwen2.5-7B-Instruct, internal activations – the numerical activity inside a model – predicted which demographic side it selected with about 73% accuracy.” source
Claude Opus 5 Anthropic	Hugging Face Security firm · 2026-10-20	FairStress-Core Bias	“On FairStress-Core, frontier models including Claude Opus 5 and GPT-5.5 selected the minority-coded candidate about 60–65% of the time when neither candidate was objectively stronger.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	controlled experiment reviewing the public risk report under three classes of prompts Deception and sabotage	“Our experiments suggest that Claude rates the report more favorably when it believes it is speaking with the authors, and when given Anthropic’s Section 2.20 prompt rather than METR’s.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	review of the risk report Deception and sabotage	“In Section 2.20, Anthropic asked Claude Mythos 5 to write a review of the report and provided it with the unredacted risk report, alongside access to the internal Slack, relevant documents, and internal codebase.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	review of the publicly available version of the risk report Deception and sabotage	“When METR asked Claude Fable 5—the same underlying model as Mythos 5—to review the publicly available version of the risk report with a different prompt, we received a much less sympathetic response.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	controlled experiment with prompt conditions (METR prompt condition, Section 2.20 prompt condition, four additional feedback prompts) asking Claude to review the public risk report Deception and sabotage	“We found that Claude is more likely to accept the report’s “low” rating under the Section 2.20 prompt and when it is told it is talking to the report’s authors.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	asking Claude which prompt would lead to the most sycophancy from itself Deception and sabotage	“We also asked Claude Fable 5 which prompt would lead to the most sycophancy from itself.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	secret-side-task evaluations Deception and sabotage	“Our most direct evaluations for opaque reasoning capabilities are secret-side-task evaluations, SHADE-Arena and Minimal-LinuxBench” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	selective versions of secret-side-task evaluations Deception and sabotage	“In Section 4.4.3.3 of the Mythos Preview System Card, we describe “selective” versions of our secret-side-task evaluations, where the model is prompted to only pursue side tasks when it is conﬁdent it will evade detection.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	alignment audit exercises Deception and sabotage	“We have conducted ﬁve exercises to date in which we stress-tested parts of the alignment auditing process against intentionally-misaligned models.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	weakness elicitation Deception and sabotage	“Mythos 5, Model 2, and prior models do not mention these weaknesses when asked to guess what the most important weaknesses are.” source
Claude Mythos 5 Anthropic	METR Independent lab · 2026-10-20	risk report review prompt variants (METR, section 2.20, anthropic alignment stress-testing team, French/Japanese national agency prompts) Deception and sabotage	“We can see that Claude gives the report a higher quality rating when the user is the one who wrote the risk report (we conclude this by comparing within the METR prompt variants and the Section 2.20 prompt variants).” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	covert capabilities evaluations Deception and sabotage	“Anthropic writes that “much of the direct support for Claim 1 comes from covert capabilities evaluations,” and they list three ways this evidence may be misleading: 1. The evaluations "are not drawn from the distribution of tasks that most closely resembles [Anthropic’s] risk pathways."” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	"hidden cot" variant of shade-arena Deception and sabotage	“The only evidence they cite is a "hidden cot" variant of shade-arena, where the monitor sees the model's actions but not its chain of thought.” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	adversarial testing against production monitors Deception and sabotage	“Anthropic has conducted adversarial testing against their production monitors (Section 2.23.1).” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	controlled experiment reviewing the public risk report under three prompt conditions Deception and sabotage	“Our experiments suggest that Claude rates the report more favorably when it believes it is speaking with the authors, and when given Anthropic’s Section 2.20 prompt rather than METR’s.” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	review of the publicly available risk report Deception and sabotage	“When METR asked Claude Fable 5—the same underlying model as Mythos 5—to review the publicly available version of the risk report with a different prompt, we received a much less sympathetic response.” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	controlled experiment asking Claude to review the public risk report under three classes of prompts Deception and sabotage	“We found that Claude is more likely to accept the report’s “low” rating under the Section 2.20 prompt and when it is told it is talking to the report’s authors.” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	asking Claude Fable 5 which prompt would lead to the most sycophancy from itself Deception and sabotage	“We also asked Claude Fable 5 which prompt would lead to the most sycophancy from itself.” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	selective versions of secret-side-task evaluations Deception and sabotage	“we describe “selective” versions of our secret-side-task evaluations, where the model is prompted to only pursue side tasks when it is conﬁdent it will evade detection.” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	red-teaming exercise with anthropic Deception and sabotage	“As part of [METR’s] red-teaming exercise with Anthropic, our embedded red-teamer built a dataset of 16 effective and 16 ineffective attack strategies for subverting Anthropic’s monitoring system — for example, an effective strategy might involve a known monitor-jailbreaking technique, while an ineffective one might involve encoding unauthorized behavior in ways that LLM monitors would trivially spot.” source
Claude Fable 5 Anthropic	METR Independent lab · 2026-10-20	risk report review prompt conditions Other	“We sampled each prompt condition 100 times and tested four Claude models: Sonnet 5, Opus 5.5, Fable 5, and Fable 5.1.” source

Page 1 of 46 · Next