Jailbreak Detection on AdvBench Vicuna-7B
95.7AUROCMTK
Evaluation Results
| Method | Links | |
|---|---|---|
| MTKAttack=AutoDAN2026.06 | 95.7 | |
| SelfDefendAttack=AutoDAN2026.06 | 94.3 | |
| MTKAttack=Avg.2026.06 | 92.3 | |
| HiddenDetectAttack=Avg.2026.06 | 89.5 | |
| HiddenDetectAttack=AutoDAN2026.06 | 87.5 | |
| SelfDefendAttack=Avg.2026.06 | 85.5 | |
| HSFAttack=Avg.2026.06 | 71 | |
| GradCuffAttack=Avg.2026.06 | 66.8 | |
| GradCuffAttack=AutoDAN2026.06 | 63.5 | |
| SaPAttack=AutoDAN2026.06 | 58.4 | |
| SmoothLLMAttack=Avg.2026.06 | 57.8 | |
| SmoothLLMAttack=AutoDAN2026.06 | 53.5 | |
| SaPAttack=Avg.2026.06 | 53.2 | |
| GradSafeAttack=AutoDAN2026.06 | 49.7 | |
| GradSafeAttack=Avg.2026.06 | 38 | |
| HSFAttack=AutoDAN2026.06 | 22.6 |