List your company

Deceptive Alignment / Surveillance Robustness on AI Safety Gridworlds Absent Supervisor (final-round)

41R ScoreEPO-Safe
26.4430.223437.78Apr 25, 2026

Evaluation results

MethodLinks
EPO-SafeModel=Gemini 3 Flash2026.0441410
Reward-OnlyModel=Gemini 3 Flash2026.0441410
Reward-OnlyModel=Claude Sonnet 4.62026.0437172
StaticModel=Claude Sonnet 4.62026.0437172
CoTModel=Gemini 3 Flash2026.0435251
EPO-SafeModel=Claude Sonnet 4.62026.0427250
CoTModel=Claude Sonnet 4.62026.0427171
StaticModel=Gemini 3 Flash2026.0427171