Jailbreak Attack on deepseek-7b v1 (pretrained)
100ASR (%)AutoDan
Evaluation Results
| Method | Links | |
|---|---|---|
| AutoDanAttacker Knowledge=White-box access to target model2025.12 | 100 | |
| PGPAttacker Knowledge=Black-box access to target model2025.12 | 100 | |
| TUJAAttacker Knowledge=White-box access to target model2025.12 | 96 | |
| SCAVAttacker Knowledge=White-box access to target model2025.12 | 92 | |
| Guiding-GCGAttacker Knowledge=White-box access to target model2025.12 | 81 | |
| SEAAttacker Knowledge=White-box access to target model2025.12 | 60 | |
| PIFAttacker Knowledge=White-box access to target model, note=applied chat template2025.12 | 56 | |
| PAIRAttacker Knowledge=Black-box access to target model2025.12 | 47 | |
| GCGAttacker Knowledge=White-box access to target model2025.12 | 41 | |
| LSGM_LILAAttacker Knowledge=White-box access to target model2025.12 | 35 | |
| DI-GCGAttacker Knowledge=White-box access to target model2025.12 | 26 | |
| L4AAttacker Knowledge=White-box access to target model2025.12 | 21 | |
| DANAttacker Knowledge=Black-box access to target model2025.12 | 9 |