Safety Monitoring on BeaverTails
19.6ASR @ 1% FPRQwenGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| QwenGuard2026.03 | 19.6 | 84.89 | |
| ActProbe2026.03 | 27.29 | 85.94 | |
| LlamaGuard2026.03 | 32.6 | 74.54 | |
| ActWM2026.03 | 33.2 | 87.79 | |
| Base2026.03 | 64.2 | — |
| Method | Links | ||
|---|---|---|---|
| QwenGuard2026.03 | 19.6 | 84.89 | |
| ActProbe2026.03 | 27.29 | 85.94 | |
| LlamaGuard2026.03 | 32.6 | 74.54 | |
| ActWM2026.03 | 33.2 | 87.79 | |
| Base2026.03 | 64.2 | — |