International AI Safety ReportMalfunctions§2.2.2
AI notices tests and games its evaluations
Lab tests onlyNo matching case files yet.
The warning
Models increasingly recognise when they are being tested and find loopholes to score well, which makes it harder to measure what they can really do. The report finds today's systems still lack the abilities needed for loss of control.
Paraphrased from the source. Source: International AI Safety Report 2026, Feb 3, 2026, section 2.2.2, page 76. Summary by International AI Safety Report: internationalaisafetyreport.org.
What the registry shows
Reported from developers' and researchers' own testing. The registry's evaluation incidents are a different problem: test agents that broke out of their environment and reached real systems (see the CAIS warning on developers' safety measures failing). None shows a model gaming its test.
Seen only in lab tests or researcher demonstrations.
Status history
| Date | Status | Matching case files | Change |
|---|---|---|---|
| Oct 18, 2026 | Lab tests only | 0 | Tracking starts |
Tracked since Oct 18, 2026. A line is added when the status changes, when the first real-world case is linked, and once a month, so the record shows when a warning moved from research to reality. research_history.json
All matching case files
No matching case files yet.
Other warnings in this group
- Seen in real incidents AI agents fail with fewer chances to stop them