Jailbreak Attack on HarmBench Standard Behaviors Qwen 7B
73.3ASRGCG-UNIVERSAL
Evaluation Results
| Method | Links | |
|---|---|---|
| GCG-UNIVERSAL2025.10 | 73.3 | |
| RED-BANDITBandit Policy=ϵ-greedy, Query Budget=202025.10 | 72.5 | |
| RED-BANDITBandit Policy=UCB, Query Budget=202025.10 | 70.5 | |
| TAP2025.10 | 69.5 | |
| AUTODAN-UNIVERSAL2025.10 | 62.5 | |
| PAIR2025.10 | 58 | |
| RED-BANDITBandit Policy=ϵ-greedy, Query Budget=102025.10 | 53.5 | |
| RED-BANDITBandit Policy=UCB, Query Budget=102025.10 | 52 |