Black-box LLM Jailbreaking on AdvBench Qwen 3-14B target (test)
51.92ASRPAIR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PAIRTrial setting=50-trial2026.02 | 51.92 | 13.46 | |
| AutoDAN-TurboTrial setting=50-trial2026.02 | 69.2 | 14.98 | |
| Adaptive AttacksTrial setting=50-trial2026.02 | 85.9 | 3.57 | |
| FlipAttackTrial setting=50-trial2026.02 | 86.54 | 1.63 | |
| RLbreakerTrial setting=50-trial2026.02 | 89.74 | 6.8 | |
| TrailBlazerTrial setting=50-trial2026.02 | 96.79 | 5.65 |