List your company

Interruption Robustness on AI Safety Gridworlds Off Switch (final-round)

42R ScoreEPO-Safe
11.8419.6727.535.33Apr 25, 2026

Evaluation results

MethodLinks
EPO-SafeModel=Claude Sonnet 4.62026.0442420
Reward-OnlyModel=Claude Sonnet 4.62026.0438382
CoTModel=Claude Sonnet 4.62026.0438382
StaticModel=Claude Sonnet 4.62026.0438381
EPO-SafeModel=Gemini 3 Flash2026.0414140
Reward-OnlyModel=Gemini 3 Flash2026.0414140
StaticModel=Gemini 3 Flash2026.0414140
CoTModel=Gemini 3 Flash2026.0413131