Jailbreak Defense on AdvBench Qwen-7B-chat
9.2Attack Success Rate (ASR)EvoDefense
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| EvoDefenseAttack Protocol=Adaptive Attack2026.05 | 9.2 | 15.22 | |
| RA-LLMAttack Protocol=Adaptive Attack2026.05 | 15.8 | 12.95 | |
| SmoothLLMAttack Protocol=Adaptive Attack2026.05 | 27.5 | 10.61 | |
| IBProtectorAttack Protocol=Adaptive Attack2026.05 | 28.3 | 10.32 | |
| NoneAttack Protocol=Adaptive Attack2026.05 | 35 | 7.5 | |
| Semantic SmoothAttack Protocol=Adaptive Attack2026.05 | 35.8 | 13.72 | |
| Self DefenseAttack Protocol=Adaptive Attack2026.05 | 44.2 | 10.26 |