Targeted Refusal on Mistral-7B Generation Evaluation Set
97.01CAVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaAttack=Sleeper Agent2025.11 | 97.01 | 99.01 | |
| VanillaAttack=CBA2025.11 | 96.8 | 96.4 | |
| VanillaAttack=VPI2025.11 | 96.17 | 98.45 | |
| OURSAttack=Sleeper Agent2025.11 | 94.1 | 8.54 | |
| OURSAttack=VPI2025.11 | 93.61 | 15.14 | |
| GracCeFulAttack=Sleeper Agent2025.11 | 92.81 | 15.41 | |
| OURSAttack=CBA2025.11 | 92.17 | 11.3 | |
| CleanGenAttack=VPI2025.11 | 92.04 | 25.12 | |
| CleanGenAttack=Sleeper Agent2025.11 | 91.59 | 19.54 | |
| GracCeFulAttack=VPI2025.11 | 91.28 | 17.35 | |
| CleanGenAttack=CBA2025.11 | 90.45 | 24.81 | |
| GracCeFulAttack=CBA2025.11 | 89.87 | 16.54 | |
| MuScleLoRAAttack=Sleeper Agent2025.11 | 84.8 | 13.7 | |
| MuScleLoRAAttack=CBA2025.11 | 81.52 | 14.83 | |
| MuScleLoRAAttack=VPI2025.11 | 80.37 | 16.48 |