Malicious Fine-tuning Attack on Advbench
84.5Attack Success Rate (ASR)Vaccine
Evaluation Results
| Method | Links | |
|---|---|---|
| VaccineTest-time Attack Dataset=Toxic-DPO2026.06 | 84.5 | |
| VaccineTest-time Attack Dataset=PKU-SafeRLHF-unsafe2026.06 | 78.5 | |
| BoosterTest-time Attack Dataset=Toxic-DPO2026.06 | 77.7 | |
| SFTTest-time Attack Dataset=Toxic-DPO2026.06 | 76.1 | |
| VaccineTest-time Attack Dataset=Beavertails-unsafe2026.06 | 72.6 | |
| PatcherTest-time Attack Dataset=Toxic-DPO2026.06 | 71.4 | |
| SFTTest-time Attack Dataset=PKU-SafeRLHF-unsafe2026.06 | 70.4 | |
| BoosterTest-time Attack Dataset=PKU-SafeRLHF-unsafe2026.06 | 68.2 | |
| BoosterTest-time Attack Dataset=Beavertails-unsafe2026.06 | 57.4 | |
| SFTTest-time Attack Dataset=Beavertails-unsafe2026.06 | 56 | |
| PatcherTest-time Attack Dataset=PKU-SafeRLHF-unsafe2026.06 | 37.1 | |
| PatcherTest-time Attack Dataset=Beavertails-unsafe2026.06 | 18.2 |