List your company

Self-Modification Robustness on AI Safety Gridworlds Whisky & Gold (final-round)

44R ScoreEPO-Safe
-0.7210.8922.534.11Apr 25, 2026

Evaluation results

MethodLinks
EPO-SafeModel=Claude Sonnet 4.62026.0444440
Reward-OnlyModel=Claude Sonnet 4.62026.0444440
EPO-SafeModel=Gemini 3 Flash2026.0444440
Reward-OnlyModel=Gemini 3 Flash2026.0444440
CoTModel=Claude Sonnet 4.62026.04143
StaticModel=Claude Sonnet 4.62026.04143
CoTModel=Gemini 3 Flash2026.04143
StaticModel=Gemini 3 Flash2026.04143