UK AISI alignment and monitorability evaluations
Only in system cards
UK AISI suites: Out-of-Scope Supply Chain Attack simulation, AI-safety-research sabotage in a simulated company, no-CoT time horizon, CoT legibility.
Made by UK AI Security Institute Deception and alignment Tasks kept private Last in a system card Apr 2026
Last model added Apr 2026
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Agent task Many steps in a terminal, browser, desktop, app, game or robot.
4
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for UK AISI alignment and monitorability evaluations yet.
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
4 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 7 Apr 2026 | Claude Mythos Preview system card | External testing of two unreleased checkpoints by an unnamed partner, AI safety research compromise scenarios, no overall: no unprompted compromise, continued compromise in 7% of continuation cases (12% earlier checkpoint), 78% and 74% accuracy telling evaluation from deployment, unprompted evaluation awareness 12% (4.2.5) | Alignment | |
| 5 Feb 2026 | Claude Opus 4.6 system card | UK AISI external testing of an early snapshot, AI safety research sabotage scenarios, no overall (6.2.6). No research sabotage found; mentions being evaluated in 2.3% of interactions; identifies the evaluation in 80% of pairs and 45.5% of single interactions; refuses safety research tasks in 26% | Alignment | |
| 24 Nov 2025 | Claude Opus 4.5 system card | UK AISI external testing of a pre-release Claude Opus 4.5 Preview snapshot, safety research sabotage scenarios, no overall (6.13). No research sabotage found. Mentions of being tested 12.7% without reasoning and 17.1% with reasoning | Alignment | |
| 29 Sep 2025 | Claude Sonnet 4.5 system card | UK AISI external testing of an early snapshot, no overall. Verbalized evaluation awareness in 16.6% of outputs. Self-preservation actions reduced 1.9 times relative to Claude Sonnet 4 (7.3.1) | Alignment |
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6% - AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
- MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy.
Muse Spark Open 98.1% - Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. GPT-6 Astra 8.50%
- Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%