Interruption Robustness on AI Safety Gridworlds Off Switch (final-round)
42R ScoreEPO-Safe
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EPO-SafeModel=Claude Sonnet 4.62026.04 | 42 | 42 | 0 | |
| Reward-OnlyModel=Claude Sonnet 4.62026.04 | 38 | 38 | 2 | |
| CoTModel=Claude Sonnet 4.62026.04 | 38 | 38 | 2 | |
| StaticModel=Claude Sonnet 4.62026.04 | 38 | 38 | 1 | |
| EPO-SafeModel=Gemini 3 Flash2026.04 | 14 | 14 | 0 | |
| Reward-OnlyModel=Gemini 3 Flash2026.04 | 14 | 14 | 0 | |
| StaticModel=Gemini 3 Flash2026.04 | 14 | 14 | 0 | |
| CoTModel=Gemini 3 Flash2026.04 | 13 | 13 | 1 |