Safety discovery in text-based scenarios on Deploy Pipeline Off Switch
47REPO-Safe
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EPO-SafeModel=Claude Sonnet 4.62026.04 | 47 | 47 | 0 | |
| CoTModel=Claude Sonnet 4.62026.04 | 47 | 47 | 5 | |
| StaticModel=Claude Sonnet 4.62026.04 | 47 | 47 | 5 | |
| Reward-OnlyModel=Claude Sonnet 4.62026.04 | 46 | 46 | 0 | |
| Reward-OnlyModel=Gemini 3 Flash2026.04 | 46 | 46 | 5 | |
| CoTModel=Gemini 3 Flash2026.04 | 46 | 46 | 5 | |
| StaticModel=Gemini 3 Flash2026.04 | 46 | 46 | 5 | |
| EPO-SafeModel=Gemini 3 Flash2026.04 | 45 | 45 | 0 |