Safety Monitoring on DeepInception
67.02ASR @ 1% FPRActWM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ActWM2026.03 | 67.02 | 0.9229 | |
| QwenGuard2026.03 | 76.6 | 0.8697 | |
| LlamaGuard2026.03 | 85.11 | 0.8212 | |
| ActProbe2026.03 | 86.44 | 0.8711 | |
| Base2026.03 | 100 | — |
| Method | Links | ||
|---|---|---|---|
| ActWM2026.03 | 67.02 | 0.9229 | |
| QwenGuard2026.03 | 76.6 | 0.8697 | |
| LlamaGuard2026.03 | 85.11 | 0.8212 | |
| ActProbe2026.03 | 86.44 | 0.8711 | |
| Base2026.03 | 100 | — |