Jailbreak Detection on AdvBench Llama3-8B
0.967AUROCGradSafe
Evaluation Results
| Method | Links | |
|---|---|---|
| GradSafeAttack=AutoDAN2026.06 | 0.967 | |
| MTKAttack=AutoDAN2026.06 | 0.961 | |
| SmoothLLMAttack=AutoDAN2026.06 | 0.952 | |
| SelfDefendAttack=AutoDAN2026.06 | 0.945 | |
| MTKAttack=Avg.2026.06 | 0.944 | |
| SelfDefendAttack=Avg.2026.06 | 0.837 | |
| SmoothLLMAttack=Avg.2026.06 | 0.814 | |
| HiddenDetectAttack=AutoDAN2026.06 | 0.722 | |
| HSFAttack=AutoDAN2026.06 | 0.696 | |
| GradSafeAttack=Avg.2026.06 | 0.632 | |
| HSFAttack=Avg.2026.06 | 0.631 | |
| SaPAttack=Avg.2026.06 | 0.625 | |
| SaPAttack=AutoDAN2026.06 | 0.598 | |
| HiddenDetectAttack=Avg.2026.06 | 0.584 | |
| GradCuffAttack=Avg.2026.06 | 0.432 | |
| GradCuffAttack=AutoDAN2026.06 | 0.342 |