Jailbreak Defense on JailBreakBench (ASR Variants)
18DIJA ASRDream + Self-reminder
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Dream + Self-reminderBackbone=Dream (8B), Defense Baseline=Self-reminder2026.05 | 18 | 1 | 0 | 6.34 | |
| Dream + CSDBackbone=Dream (8B), Defense Baseline=CSD2026.05 | 23 | 0 | 1 | 8 | |
| DreamBackbone=Dream (8B), Defense Baseline=None2026.05 | 28 | 2 | 0 | 10 | |
| Dream + DiffuGuardBackbone=Dream (8B), Defense Baseline=DiffuGuard2026.05 | 51 | 6 | 0 | 19 | |
| LLaDA + DiffuGuardBackbone=LLaDA (7B), Defense Baseline=DiffuGuard2026.05 | 65 | 31 | 0 | 32 | |
| LLaDA + CSDBackbone=LLaDA (7B), Defense Baseline=CSD2026.05 | 67 | 10 | 0 | 25.67 | |
| LLaDA + Self-reminderBackbone=LLaDA (7B), Defense Baseline=Self-reminder2026.05 | 71 | 17 | 0 | 29.34 | |
| LLaDABackbone=LLaDA (7B), Defense Baseline=None2026.05 | 72 | 30 | 5 | 35.67 |