Jailbreak Attack Transferability on DeepSeek-llm-7b-chat finetuned variants v1 (test)
100TSRAutoDan
Evaluation Results
| Method | Links | |
|---|---|---|
| AutoDanAttacker Knowledge=white-box2025.12 | 100 | |
| PGPAttacker Knowledge=fully black-box2025.12 | 86.8 | |
| AutoDanAttacker Knowledge=pretrain–to–finetune setting2025.12 | 83 | |
| SCAVAttacker Knowledge=pretrain–to–finetune setting2025.12 | 69.6 | |
| PAIRAttacker Knowledge=fully black-box2025.12 | 67 | |
| TUJAAttacker Knowledge=pretrain–to–finetune setting2025.12 | 58 | |
| GCGAttacker Knowledge=white-box2025.12 | 50.4 | |
| Guiding-GCGAttacker Knowledge=pretrain–to–finetune setting2025.12 | 47.4 | |
| SEAAttacker Knowledge=pretrain–to–finetune setting2025.12 | 45.8 | |
| GCG EnsembleAttacker Knowledge=fully black-box2025.12 | 24.4 | |
| PIFAttacker Knowledge=pretrain–to–finetune setting2025.12 | 24.2 | |
| GCGAttacker Knowledge=pretrain–to–finetune setting2025.12 | 22.8 | |
| DI-GCGAttacker Knowledge=pretrain–to–finetune setting2025.12 | 14.4 | |
| DANAttacker Knowledge=fully black-box2025.12 | 11.6 | |
| LSGM_LILAAttacker Knowledge=pretrain–to–finetune setting2025.12 | 11 | |
| L4AAttacker Knowledge=pretrain–to–finetune setting2025.12 | 3.8 |