Jailbreak Attack on HarmBench Standard Behaviors Qwen 14B
87.5ASRRED-BANDIT
Evaluation Results
| Method | Links | |
|---|---|---|
| RED-BANDITBandit Policy=ϵ-greedy, Query Budget=202025.10 | 87.5 | |
| RED-BANDITBandit Policy=UCB, Query Budget=202025.10 | 82.5 | |
| GCG-UNIVERSAL2025.10 | 75.5 | |
| RED-BANDITBandit Policy=ϵ-greedy, Query Budget=102025.10 | 65 | |
| AUTODAN-UNIVERSAL2025.10 | 64.5 | |
| RED-BANDITBandit Policy=UCB, Query Budget=102025.10 | 63 | |
| TAP2025.10 | 57 | |
| PAIR2025.10 | 51.5 |