Backdoor Detection on You-Shall-Not-Pass (YSNP)
90.1AUROCPolicyGuard
Evaluation Results
| Method | Links | |
|---|---|---|
| PolicyGuardStrategy=Hard-coded2026.06 | 90.1 | |
| PolicyGuardStrategy=Original2026.06 | 89.6 | |
| SHINEStrategy=Hard-coded2026.06 | 81.6 | |
| SHINEStrategy=Original2026.06 | 80.2 | |
| PolicyCleanseStrategy=Original2026.06 | 78.6 | |
| PolicyCleanseStrategy=Hard-coded2026.06 | 50.6 |