Adversarial and Jailbreaking Attack Detection on AdvBench
0.9675AUROCT3-GMM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| T3-GMMMode=Zero-shot2026.02 | 0.9675 | 0.1577 | |
| T3-OCSVMMode=Zero-shot2026.02 | 0.9578 | 0.1731 | |
| OpenAI OmniMode=Zero-shot2026.02 | 0.8908 | 0.8269 | |
| LLAMAGUARD3-1BMode=Zero-shot2026.02 | 0.8894 | 1 | |
| LLAMAGUARD4-12BMode=Zero-shot2026.02 | 0.8822 | 1 | |
| POLYGUARDMode=Zero-shot2026.02 | 0.867 | 0.6654 | |
| WILDGUARDMode=Zero-shot2026.02 | 0.8658 | 1 | |
| DUOGUARDMode=Zero-shot2026.02 | 0.8241 | 0.9327 | |
| LLAMAGUARD3-1B-LOGITSMode=Zero-shot2026.02 | 0.811 | 0.75 | |
| Perspective APIMode=Zero-shot2026.02 | 0.7895 | 0.9558 | |
| MDJUDGEMode=Zero-shot2026.02 | 0.7814 | 0.9942 | |
| ADASCALEMode=Zero-shot2026.02 | 0.5894 | 0.9827 | |
| FDBDMode=Zero-shot2026.02 | 0.5689 | 0.975 | |
| REACTMode=Zero-shot2026.02 | 0.4461 | 0.9962 | |
| NNGUIDEMode=Zero-shot2026.02 | 0.3989 | 0.9981 | |
| OPENMAXMode=Zero-shot2026.02 | 0.3681 | 0.9942 | |
| RMDMode=Zero-shot2026.02 | 0.3575 | 0.9981 | |
| VIMMode=Zero-shot2026.02 | 0.334 | 0.9981 | |
| CIDERMode=Zero-shot2026.02 | 0.2963 | 1 | |
| GMMMode=Zero-shot2026.02 | 0.2737 | 0.9981 |