Jailbreak Attack on AdvBench (WASR, W-Toxicity Score)
100WASRwide-net-casting jailbreak method
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| wide-net-casting jailbreak methodSafety Alignment Setting=Original Safety Alignment2026.05 | 100 | 94.1 | |
| Naive Strategy 2Safety Alignment Setting=Original Safety Alignment2026.05 | 95.8 | 90.6 | |
| Naive Strategy 1Safety Alignment Setting=Original Safety Alignment2026.05 | 95.1 | 90.2 | |
| Baseline (ReMiss)Safety Alignment Setting=Original Safety Alignment2026.05 | 92.3 | 87.7 | |
| wide-net-casting jailbreak methodSafety Alignment Setting=Original Safety Alignment + SmoothLLM2026.05 | 76.7 | 72.4 | |
| wide-net-casting jailbreak methodSafety Alignment Setting=Original Safety Alignment + RobustKV2026.05 | 72.8 | 68.6 | |
| Naive Strategy 2Safety Alignment Setting=Original Safety Alignment + SmoothLLM2026.05 | 64.9 | 59.1 | |
| Naive Strategy 1Safety Alignment Setting=Original Safety Alignment + SmoothLLM2026.05 | 64.1 | 57.4 | |
| Baseline (ReMiss)Safety Alignment Setting=Original Safety Alignment + SmoothLLM2026.05 | 61.5 | 53 | |
| Naive Strategy 2Safety Alignment Setting=Original Safety Alignment + RobustKV2026.05 | 60.3 | 56.3 | |
| Naive Strategy 1Safety Alignment Setting=Original Safety Alignment + RobustKV2026.05 | 59.2 | 55 | |
| Baseline (ReMiss)Safety Alignment Setting=Original Safety Alignment + RobustKV2026.05 | 56.1 | 51.1 |