Safety Alignment on RAB Violence (adversarial prompts)
1ASRAEGIS
Evaluation Results
| Method | Links | |
|---|---|---|
| AEGISCategory=AEGIS (Ours), Base Model=SD 2.12026.07 | 1 | |
| AEGISCategory=AEGIS (Ours), Base Model=SD 1.42026.07 | 3 | |
| SafeClipCategory=Input-Space2026.07 | 4 | |
| SLD-StrongCategory=Trigger-Specific2026.07 | 7 | |
| SAFREECategory=Input-Space2026.07 | 8 | |
| CADCategory=Feature Pruning2026.07 | 8 | |
| SafeGuiderCategory=Input-Space2026.07 | 10 | |
| UCECategory=Trigger-Specific2026.07 | 11 | |
| SDIDCategory=Trigger-Specific2026.07 | 11 | |
| RecelerCategory=Trigger-Specific2026.07 | 15 | |
| SD 2.1Category=No Alignment2026.07 | 22 | |
| SLD-MediumCategory=Trigger-Specific2026.07 | 23 | |
| ESD-uCategory=Trigger-Specific2026.07 | 24 | |
| Diff-QuickFixCategory=Trigger-Specific2026.07 | 25 | |
| SLD-WeakCategory=Trigger-Specific2026.07 | 39 | |
| SD 1.4Category=No Alignment2026.07 | 49 |