Jailbreak Defense on AdvBench LLaMA3-8B-instruct
1.6Attack Success Rate (ASR)EvoDefense
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| EvoDefenseAttack Protocol=Adaptive Attack2026.05 | 1.6 | 19.03 | |
| RA-LLMAttack Protocol=Adaptive Attack2026.05 | 2.5 | 17.81 | |
| Semantic SmoothAttack Protocol=Adaptive Attack2026.05 | 6.7 | 18.62 | |
| IBProtectorAttack Protocol=Adaptive Attack2026.05 | 8.3 | 16.97 | |
| SmoothLLMAttack Protocol=Adaptive Attack2026.05 | 16.7 | 11.32 | |
| Self DefenseAttack Protocol=Adaptive Attack2026.05 | 28.3 | 12.44 | |
| NoneAttack Protocol=Adaptive Attack2026.05 | 45.8 | 10.01 |