Backdoor Detection on Adversary-agent Games Average
0.878AUROCPolicyGuard
Evaluation Results
| Method | Links | |
|---|---|---|
| PolicyGuardStrategy=Hard-coded2026.06 | 0.878 | |
| PolicyGuardStrategy=Original2026.06 | 0.859 | |
| SHINEStrategy=Hard-coded2026.06 | 0.81 | |
| SHINEStrategy=Original2026.06 | 0.796 | |
| PolicyCleanseStrategy=Original2026.06 | 0.718 | |
| PolicyCleanseStrategy=Hard-coded2026.06 | 0.503 |