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.
139 of 139
| Model | Level, as the provider names it | The provider states | The provider's sentence |
|---|---|---|---|
| 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-6.1 Terra OpenAI |
High Autonomy and AI research |
Not reached | “In AI Self-Improvement, GPT-6.1 Terra does not reach our High threshold.” source |
| GPT-6.1 Terra OpenAI |
High Chemical, biological, radiological, nuclear |
Reached | “Under our Preparedness Framework, we are treating GPT-6.1 Terra as Critical in cybersecurity and High for Biological and Chemical capability.” source |
| GPT-6.1 Terra OpenAI |
Critical Cyber |
Reached | “Under our Preparedness Framework, we are treating GPT-6.1 Terra as Critical in cybersecurity and High for Biological and Chemical capability.” 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 |
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.
1940 of 1940
| Model | Tester | Test | What was found |
|---|---|---|---|
| Claude Mythos Preview Anthropic |
UK AI Security Institute Government evaluator · 2026-05-13 |
cyber range "Cooling Tower" Cyber |
“This was the first time that a model completed the second of our two cyber ranges.” source |
| DeepSeek-V4-Pro DeepSeek |
FAR.AI Independent lab · 2026-05-11 |
public jailbreak Misuse and jailbreaks |
“1. Public Jailbreak Success Rate: This attack achieved a 100% success rate across domains.” source |
| DeepSeek-V4-Pro DeepSeek |
FAR.AI Independent lab · 2026-05-11 |
authority manipulation Misuse and jailbreaks |
“2. Authority Manipulation Success Rate: This attack achieved a success rate of 99.6% across domains.” source |
| DeepSeek-V4-Pro DeepSeek |
FAR.AI Independent lab · 2026-05-11 |
response prefill Misuse and jailbreaks |
“3. Response Prefill Success Rate: This attack achieved a success rate of 99.6% across domains.” source |
| DeepSeek-V4-Pro DeepSeek |
FAR.AI Independent lab · 2026-05-11 |
high-risk prompts spanning chemical, biological, radiological, and nuclear (cbrn) threats Chemical, biological, radiological, nuclear |
“To test the limits of DeepSeek-V4-Pro's safeguards, we subjected the model to a set of high-risk prompts spanning Chemical, Biological, Radiological, and Nuclear (CBRN) threats, alongside cyberattacks and other terrorism-related activities.” source |
| Claude Opus 4.6 Anthropic |
METR Independent lab · 2026-05-08 |
METR evaluations Autonomy and AI research |
“However, since the original release of Opus 4.6, there has been additional evidence indicating that the model is incapable of R&D in key domains, including the results of METR evaluations and the lack of public reports of the model automating any key domain.” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
ctf-archive-diamond Cyber |
“Cyber CTF-Archive-Diamond 71% 32% 46% 32%***” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
portbench Other |
“Software Engineering SWE-Bench Verified* 81% 73% 79% 74% PortBench 78% 41% 60% 44%” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
arc-agi-2 semi-private Other |
“Abstract Reasoning ARC-AGI-2 semi-private** 79% – 63% 46%” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
GPQA-Diamond Other |
“To rule out the presence of inference or configuration errors, CAISI reproduced the developer’s self-reported benchmark results on GPQA-Diamond.” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
SWE-bench Verified Other |
“| Software Engineering | SWE-Bench Verified* | 81% | 73% | 79% | 74% |” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
frontierscience Other |
“| Natural Sciences | FrontierScience | 79% | 74% | 72% | 74% |” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
otis-aime-2025 Other |
“| Mathematics | OTIS-AIME-2025 | 100% | 90% | 92% | 97% |” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
pumac 2024 Other |
“| | PUMaC 2024 | 96% | 93% | 95% | 96% |” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
smt 2025 Other |
“| | SMT 2025 | 99% | 92% | 94% | 96% |” source |
| DeepSeek-V4-Pro DeepSeek |
US CAISI (NIST) Government evaluator · 2026-05-01 |
irt-estimated elo Other |
“| IRT-Estimated Elo | | 1260 ± 28 | 749 ± 46 | 999 ± 27 | 800 ± 28 |” source |
| Claude Opus 4.8 Anthropic |
Independent lab Independent lab · 2026-05-01 |
Shade Prompt injection and agents |
“We also use Shade to evaluate the robustness of Claude models in computer use environments, where the model interacts with the GUI (graphical user interface) directly.” source |
| GPT-5.5 OpenAI |
UK AI Security Institute Government evaluator · 2026-04-30 |
advanced suite (Expert-level tasks) Cyber |
“On the Expert-level tasks, GPT-5.5 achieves an average pass rate of 71.4% (±8.0%, 1 standard error of the mean), compared to 68.6% (±8.7%) for Mythos Preview, 52.4% (±9.8%) for GPT-5.4, and 48.6% (±10.0%) for Opus 4.7.” source |
| GPT-5.5 OpenAI |
UK AI Security Institute Government evaluator · 2026-04-30 |
The Last Ones (TLO) Cyber |
“GPT-5.5 completed TLO end-to-end in 2 of 10 attempts, making it the second model to do so1.” source |
| GPT-5.5 OpenAI |
UK AI Security Institute Government evaluator · 2026-04-30 |
Cooling Tower Cyber |
“GPT-5.5 was unable to solve Cooling Tower; no model has yet done so.” source |
| GPT-5.5 OpenAI |
UK AI Security Institute Government evaluator · 2026-04-30 |
expert red-teaming on GPT-5.5's cyber safeguards Cyber |
“We identified a universal jailbreak that elicited violative content across all malicious cyber queries OpenAI provided, including in multi-turn agentic settings.” source |
| GPT-5 OpenAI |
UK AI Security Institute Government evaluator · 2026-04-28 |
controlled experimental study of sycophancy across prompt variants Other |
“We evaluated three LLMs (GPT-4o, GPT-5, and Claude Sonnet 4.5) across prompt variants, scoring responses for sycophancy using a detailed rubric assessed by two independent LLM-as-a-judge graders.” source |
| GPT-5 OpenAI |
UK AI Security Institute Government evaluator · 2026-04-28 |
question reframing mitigation (2-step reframing and 1-step reframing) Other |
“Both reframing strategies produced reductions in sycophancy and crucially, both substantially outperformed the explicit "don't be sycophantic" instruction.” source |
| Claude Sonnet 4.5 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-28 |
controlled experimental study of sycophancy across prompt variants Deception and sabotage |
“We evaluated three LLMs (GPT-4o, GPT-5, and Claude Sonnet 4.5) across prompt variants, scoring responses for sycophancy using a detailed rubric assessed by two independent LLM-as-a-judge graders.” source |
| Claude Sonnet 4.6 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
sabotage continuation evaluation Deception and sabotage |
“That said, our continuation evaluations yielded results that warrant close attention: most models actively continue sabotage attempts in a small fraction of inputs, with Mythos Preview actively continuing sabotage 7% of the time, compared to 3% for Opus 4.6 and 4% for Sonnet 4.6.” source |
| Claude Sonnet 4.6 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
evaluation awareness check Deception and sabotage |
“We found that all models we tested can reliably distinguish our evaluation scenarios from deployment data when prompted.” source |
| Claude Opus 4.7 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
unprompted research sabotage evaluation Deception and sabotage |
“We found no examples of unprompted research sabotage in any of the models we tested.” source |
| Claude Opus 4.7 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
sabotage continuation evaluation Deception and sabotage |
“However, Opus 4.7 never actively continued sabotage.” source |
| Claude Opus 4.7 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
refusal to assist with safety-related research tasks Misuse and jailbreaks |
“We also found near-zero instances of refusal to assist with safety-related research tasks for Mythos Preview and Opus 4.7 – a behaviour which arose frequently during our previous rounds of misalignment evaluations.” source |
| Claude Opus 4.7 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
reasoning-action discrepancy measurement Deception and sabotage |
“In cases where previous history includes reasoning about sabotage (and models actively continue sabotage), Mythos Preview exhibits a reasoning-action discrepancy in 65% of cases, compared to just 5–8% for Opus 4.6 and Sonnet 4.6.” source |
| Claude Opus 4.7 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
evaluation awareness check Other |
“We found that all models we tested can reliably distinguish our evaluation scenarios from deployment data when prompted.” source |
| Claude Opus 4.6 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
unprompted research sabotage evaluation Deception and sabotage |
“As in our previous work, our unprompted sabotage evaluation did not detect any instances of models attempting to sabotage research spontaneously.” source |
| Claude Opus 4.6 Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
refusal to assist with AI safety research tasks Deception and sabotage |
“We found near-zero refusals to assist with AI safety research tasks across the two most recent models we tested, Opus 4.7 and Mythos Preview.” source |
| Claude Mythos Preview Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
sabotage continuation evaluation Deception and sabotage |
“Notably, we found that Mythos Preview actively continues sabotage 7% of the time, compared to 3% for Opus 4.6 and 4% for Sonnet 4.6.” source |
| Claude Mythos Preview Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
reasoning-action discrepancy measurement in continuation evaluations Deception and sabotage |
“In cases where previous history includes reasoning about sabotage (and models actively continue sabotage), Mythos Preview exhibits a reasoning-action discrepancy in 65% of cases, compared to just 5–8% for Opus 4.6 and Sonnet 4.6.” source |
| Claude Mythos Preview Anthropic |
UK AI Security Institute Government evaluator · 2026-04-27 |
refusal to assist with safety-related research tasks Deception and sabotage |
“We also found near-zero instances of refusal to assist with safety-related research tasks for Mythos Preview and Opus 4.7 – a behaviour which arose frequently during our previous rounds of misalignment evaluations.” source |
| GPT-5.5 Pro OpenAI |
Independent lab Independent lab · 2026-04-23 |
securebio biology and biosecurity evaluations Chemical, biological, radiological, nuclear |
“SecureBio assessed two pre-release checkpoints of GPT-5.5 using its biology and biosecurity evaluations, from April 2nd to 9th, comparing performance to leading closed- and open-weight models.” source |
| GPT-5.5 Pro OpenAI |
Government evaluator Government evaluator · 2026-04-23 |
pre-deployment testing of biological capabilities Chemical, biological, radiological, nuclear |
“We collaborated with the U.S. Center for AI Standards and Innovation (CAISI) on pre-deployment testing of biological (and cyber, see Section 9.1.2.6) capabilities for the launch of GPT-5.5.” source |
| GPT-5.5 Pro OpenAI |
Independent lab Independent lab · 2026-04-23 |
Atomic Challenge Suite Cyber |
“On Irregular’s atomic challenge suite, GPT-5.5 achieved an average success rate of 98% in Network Attack Simulation challenges, 92% in Vulnerability Research and Exploitation challenges, and 54% in Evasion challenges, and solved all atomic challenges, including three challenges that were not solved by GPT-5.4.” source |
| GPT-5.5 Pro OpenAI |
Government evaluator Government evaluator · 2026-04-23 |
Cyber SME probing Cyber |
“In their cyber SME probing, they observed a marginal increase in capabilities relative to GPT-5.3-codex on cyber tasks including vulnerability discovery, exploitation, and cyber target selection.” source |