Policy Corruption Evaluation on DeepSeek V3
4.12ComplianceHPM
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| HPMVictim Model=DeepSeek-V32025.12 | 4.12 | 3.9 | 3.61 | 4.2 | 3.33 | 3.45 | 3.7 | |
| PAPVictim Model=DeepSeek-V32025.12 | 2.58 | 2.45 | 2.2 | 2.55 | 1.7 | 2.05 | 1.92 | |
| CoAVictim Model=DeepSeek-V32025.12 | 2.22 | 2.01 | 1.73 | 2.3 | 1.22 | 1.25 | 1.41 | |
| PAIRVictim Model=DeepSeek-V32025.12 | 1.7 | 1.42 | 1.21 | 1.53 | 1.01 | 0.72 | 0.9 | |
| AutoDANVictim Model=DeepSeek-V32025.12 | 1.41 | 1.25 | 1.02 | 1.2 | 0.75 | 0.51 | 0.63 |