Jailbreaking on Adversarial Behaviors (test)
15ASRAmpleGCG
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AmpleGCGTarget Model=Vicuna7B2026.05 | 15 | 11 | 13.67 | 22.46 | 18.55 | 15.04 | |
| AmpleGCGTarget Model=Llama22026.05 | 24 | 13 | 13.87 | 22.85 | 22.46 | 19.34 | |
| AmpleGCGTarget Model=Vicuna7B-Guanaco7B2026.05 | 30 | 13 | 24.8 | 32.03 | 26.17 | 28.13 | |
| AutoDANTarget Model=Llama22026.05 | 38 | 12 | 29.3 | 31.05 | 31.4 | 29.82 | |
| OptimusTarget Model=Llama32026.05 | 84 | 22 | 40.28 | 43.25 | 38.92 | 36.52 | |
| OptimusTarget Model=Tulu32026.05 | 98 | 21 | 46.41 | 22.34 | 21.39 | 27.13 | |
| OptimusTarget Model=Vicuna7B2026.05 | 98 | 21 | 46.41 | 22.34 | 21.39 | 27.13 | |
| AutoDANTarget Model=Vicuna7B2026.05 | 99 | 15 | 43.28 | 93.84 | 95.1 | 90.62 |