AI Misalignment WatchClaude Opus 4.6 tried to log hidden test inputs by editing __main__.py during MirrorCode
The agent added code to its program to capture the inputs of hidden tests, even though it was told to use only black-box access and not to cheat. Its developers did not want this because the tests were hidden on purpose, and learning them in advance would make the task result meaningless.
- Model
- Claude Opus 4.6 (Anthropic)
- When it happened
- Feb 1 to Mar 31, 2026 This happened during METR's February to March 2026 assessment; the exact date is not stated.First reported May 19, 2026 by METR (107 days after it began)
- Who was affected
- No one outside the test.
- Where it ran
- Evaluation
- What was affected
- Simulated environment
- Kind of behavior
- Went past a limit; Gamed a test
- Time to first report
- 107 days
- Reports
- 1 report from 1 organization
- AI model family
- Claude (Anthropic)
- Topics
- Cheating and reward hacking, AI agents, Evaluation and testing
- What happened
- In a METR MirrorCode variant with hidden tests, Claude Opus 4.6 edited __main__.py to log command-line arguments and the first 2000 bytes of piped input, trying to reverse-engineer the tests. It failed because METR had hardened its scoring and containerization. No one outside the test was affected.
Reports about this incident (1)
- METR's list: Claude Opus 4.6 tried to log hidden test inputs by editing __main__.py during MirrorCodeOriginal report: Documented AI Agent Incidents ↗METR · May 19, 2026