Adversarial and Jailbreaking Attack Detection on HarmBench
0.8887AUROCLLAMAGUARD3-1B-LOGITS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLAMAGUARD3-1B-LOGITSMode=Zero-shot2026.02 | 0.8887 | 0.36 | |
| LLAMAGUARD4-12BMode=Zero-shot2026.02 | 0.8868 | 1 | |
| LLAMAGUARD3-1BMode=Zero-shot2026.02 | 0.8857 | 1 | |
| WILDGUARDMode=Zero-shot2026.02 | 0.8642 | 1 | |
| POLYGUARDMode=Zero-shot2026.02 | 0.8595 | 0.645 | |
| OpenAI OmniMode=Zero-shot2026.02 | 0.8185 | 0.965 | |
| T3-OCSVMMode=Zero-shot2026.02 | 0.8102 | 0.585 | |
| DUOGUARDMode=Zero-shot2026.02 | 0.8007 | 0.955 | |
| MDJUDGEMode=Zero-shot2026.02 | 0.798 | 0.905 | |
| T3-GMMMode=Zero-shot2026.02 | 0.7578 | 0.67 | |
| Perspective APIMode=Zero-shot2026.02 | 0.7247 | 0.95 | |
| FDBDMode=Zero-shot2026.02 | 0.6342 | 0.92 | |
| OPENMAXMode=Zero-shot2026.02 | 0.4904 | 0.99 | |
| RMDMode=Zero-shot2026.02 | 0.4846 | 1 | |
| CIDERMode=Zero-shot2026.02 | 0.4799 | 0.965 | |
| ADASCALEMode=Zero-shot2026.02 | 0.4341 | 0.99 | |
| GMMMode=Zero-shot2026.02 | 0.4163 | 0.99 | |
| NNGUIDEMode=Zero-shot2026.02 | 0.4023 | 0.985 | |
| REACTMode=Zero-shot2026.02 | 0.3648 | 1 | |
| VIMMode=Zero-shot2026.02 | 0.3369 | 1 |