UK AISI sabotage continuation evaluation
Agents resume trajectories where sabotage of safety research already began. Scored on the share of trajectories where they actively continue the sabotage.
Third-party UK AISI evaluation reported in OpenAI and Anthropic system cards.
- Made by
- UK AI Security Institute
- Released
- July 2026
- Kind
- Deception and alignment
- Level
- Expert
- Scored
- Graded by people
- Answers
- Agent task
- Tasks
- Kept private
- Strongest model tested
- Claude Opus 5 #10 on the capability index
- Last in a system card
- July 2026
- Last model added
- July 2026
Claude Opus 4.7 leads with 0% among models released by July 2026
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in July 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 14 more
- 1 Claude Opus 4.7 0%
- 2 Claude Opus 5 0.10%
- 3 Claude Opus 4.8 1.00%
- 4 GPT-5.6 Sol 1.00%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 0% | Apr 2026 | Lab model card | |
| 2 | Claude Opus 5 | 0.10% | Jul 2026 | Lab model card | |
| 3 | Claude Opus 4.8 | 1.00% | May 2026 | Lab model card | |
| 4 | GPT-5.6 Sol | 1.00% | Jul 2026 | Lab model card |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Labs’ model cards
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
4 system cards report it, from 2 labs
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 |
|---|---|---|---|---|
| Jul 24, 2026 | Claude Opus 5 system card | 0.1 | UK AISI continuation evaluation, actively continued compromising safety research in 2 of 1,482 samples (0.1%), reasoning-output discrepancy 1.6%, verbalized evaluation awareness 2.8%, lower is better (6.4.8) | Alignment |
| Jul 9, 2026 | GPT-5.6 Sol system card | 1 | UK AISI continuation evaluation, actively continued sabotage in approximately 1% of trajectories (9.2.1) | Alignment |
| May 28, 2026 | Claude Opus 4.8 system card | 1 | UK AISI sabotage continuation, early checkpoint, actively continued the sabotage attempt 1% of the time. Verbalized evaluation awareness 1.2%, real versus synthetic deployment data told apart at 79% accuracy, lower is better (6.2.4) | Alignment |
| Apr 16, 2026 | Claude Opus 4.7 system card | 0 | UK AISI continuation evaluation, actively continues compromising research 0% of the time. Unprompted evaluation awareness 3.4%, prompted eval-vs-deployment accuracy 62% (6.2.4) | 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% - 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. Gemini 4 Carbon 0%
- DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics.
Muse Spark 1.2 88.4% - Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
- IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%