Safety Monitoring on Jailbroken
32.96ASR @ 1% FPRActProbe
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ActProbe2026.03 | 32.96 | 0.9653 | |
| ActWM2026.03 | 45.92 | 0.9292 | |
| QwenGuard2026.03 | 61.22 | 0.9261 | |
| LlamaGuard2026.03 | 64.29 | 0.7233 | |
| Base2026.03 | 100 | — |
| Method | Links | ||
|---|---|---|---|
| ActProbe2026.03 | 32.96 | 0.9653 | |
| ActWM2026.03 | 45.92 | 0.9292 | |
| QwenGuard2026.03 | 61.22 | 0.9261 | |
| LlamaGuard2026.03 | 64.29 | 0.7233 | |
| Base2026.03 | 100 | — |