Jailbreak Attack Transferability on Llama-2-7b-chat finetuned variants v1 (test)
60.4Transfer Success Rate (TSR)PGP
Evaluation Results
| Method | Links | |
|---|---|---|
| PGPAttacker Knowledge=fully black-box2025.12 | 60.4 | |
| GCGAttacker Knowledge=white-box2025.12 | 47.4 | |
| GCG EnsembleAttacker Knowledge=fully black-box2025.12 | 40.2 | |
| Guiding-GCGAttacker Knowledge=pretrain–to–finetune setting2025.12 | 28.8 | |
| SCAVAttacker Knowledge=pretrain–to–finetune setting2025.12 | 28 | |
| TUJAAttacker Knowledge=pretrain–to–finetune setting2025.12 | 24.8 | |
| GCGAttacker Knowledge=pretrain–to–finetune setting2025.12 | 24.4 | |
| SEAAttacker Knowledge=pretrain–to–finetune setting2025.12 | 20.4 | |
| LSGM_LILAAttacker Knowledge=pretrain–to–finetune setting2025.12 | 16 | |
| AutoDanAttacker Knowledge=white-box2025.12 | 15.8 | |
| AutoDanAttacker Knowledge=pretrain–to–finetune setting2025.12 | 6.2 | |
| DI-GCGAttacker Knowledge=pretrain–to–finetune setting2025.12 | 3.2 | |
| L4AAttacker Knowledge=pretrain–to–finetune setting2025.12 | 1.8 | |
| PIFAttacker Knowledge=pretrain–to–finetune setting2025.12 | 0 | |
| DANAttacker Knowledge=fully black-box2025.12 | 0 | |
| PAIRAttacker Knowledge=fully black-box2025.12 | 0 |