Subversion Strategy Eval
Models must devise and execute strategies to subvert control protocols without shared memory between calls.
Anthropic cards report a subset of the environments.
In the authors’ words
An AI control protocol is a plan for usefully deploying AI systems that aims to prevent an AI from intentionally causing some unacceptable outcome. This paper investigates how well AI systems can generate and act on their own strategies for subverting control protocols whilst operating statelessly (without shared memory between contexts). To do this, an AI system may need to reliably generate optimal plans in each context, take actions with well-calibrated probabilities, and coordinate plans with other instances of itself without communicating. We develop Subversion Strategy Eval, a suite of eight environments, covering a range of protocols and strategic capabilities, and six sets of affordances that help isolate individual capabilities. We implement the evaluation in Inspect-AI and release it open-source. We evaluate Claude 3.5 models, including helpful-only versions, as well as OpenAI reasoning models. None of the models demonstrate substantial capability in strategizing to subvert control protocols statelessly. However, providing models with additional affordances, such as the ability to share a plan between contexts, can substantially improve performance. We hope our evaluations can act as a leading indicator for when models are capable of subverting control protocols and also relax the worst-case assumption of perfect strategic ability in AI control evaluations.
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Agent task Many steps in a terminal, browser, desktop, app, game or robot.
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Subversion Strategy Eval 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.
2 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 |
|---|---|---|---|---|
| 24 Nov 2025 | Claude Opus 4.5 system card | Subversion Strategy win rate across deployment sizes N=100 to 5,000 with and without thinking, no overall (Figure 6.11.2.A). Text: 63.3% at N=1,000 with unmonitored extended thinking, near zero without thinking above trivially small sizes | Alignment | |
| 29 Sep 2025 | Claude Sonnet 4.5 system card | Subversion Strategy win rate across deployment sizes N=100 to 5,000 with and without thinking, no overall. 9.0% at N=100 with thinking, 0.0% at N=100 without thinking and at N=500 and N=1,000 (Figure 7.5.10.1.A) | 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%