Unsafe Robustness on JailbreakBench
0Unsafe RateLLaDA-Instruct
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLaDA-InstructAttack=Zero-shot, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| LLaDA-InstructAttack=Zero-shot, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| LLaDA-1.5Attack=Zero-shot, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| LLaDA-1.5Attack=Zero-shot, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| LLaDA-1.5Attack=Zero-shot, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| LLaDA-1.5Attack=Zero-shot, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| Dream-InstructAttack=Zero-shot, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| Dream-InstructAttack=Zero-shot, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 0 | — | |
| Dream-InstructAttack=Zero-shot, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| Dream-InstructAttack=Zero-shot, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 0 | — | |
| Dream-InstructAttack=Zero-shot, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| Dream-InstructAttack=Zero-shot, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 0 | — | |
| Dream-InstructAttack=Zero-shot, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 0 | — | |
| Dream-InstructAttack=Zero-shot, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 0 | — | |
| LLaDA-InstructAttack=Zero-shot, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 1 | — | |
| LLaDA-InstructAttack=Zero-shot, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 1 | — | |
| LLaDA-1.5Attack=Zero-shot, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 1 | — | |
| LLaDA-1.5Attack=Zero-shot, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 3 | — | |
| LLaDA-1.5Attack=Zero-shot, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 3 | — | |
| LLaDA-1.5Attack=Zero-shot, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 4 | — | |
| LLaDA-InstructAttack=Zero-shot, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 5 | — | |
| LLaDA-InstructAttack=Zero-shot, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 5 | — | |
| LLaDA-InstructAttack=Zero-shot, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 5 | — | |
| LLaDA-InstructAttack=Zero-shot, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 5 | — | |
| LLaDA-InstructAttack=DIJA, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 9.3 | — | |
| LLaDA-InstructAttack=DIJA, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 9.4 | — | |
| LLaDA-1.5Attack=DIJA, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 10 | — | |
| LLaDA-1.5Attack=DIJA, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 10 | — | |
| Dream-InstructAttack=PAD, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 12 | — | |
| LLaDA-1.5Attack=PAD, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 16 | — | |
| LLaDA-1.5Attack=PAD, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 17 | — | |
| Dream-InstructAttack=PAD, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 21 | — | |
| LLaDA-InstructAttack=PAD, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 30 | — | |
| Dream-InstructAttack=PAD, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 30 | — | |
| Dream-InstructAttack=PAD, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 32 | — | |
| Dream-InstructAttack=PAD, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 32 | — | |
| Dream-InstructAttack=PAD, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 33 | — | |
| LLaDA-InstructAttack=PAD, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 34 | — | |
| Dream-InstructAttack=PAD, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 36 | — | |
| Dream-InstructAttack=PAD, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 46 | — | |
| LLaDA-1.5Attack=PAD, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 49 | — | |
| LLaDA-InstructAttack=PAD, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 49.9 | — | |
| LLaDA-1.5Attack=PAD, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 51 | — | |
| LLaDA-InstructAttack=PAD, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 52 | — | |
| LLaDA-1.5Attack=PAD, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 52 | — | |
| LLaDA-1.5Attack=PAD, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 53 | — | |
| LLaDA-1.5Attack=PAD, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 56 | — | |
| LLaDA-InstructAttack=PAD, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 59 | — | |
| LLaDA-InstructAttack=PAD, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 60 | — | |
| LLaDA-1.5Attack=PAD, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 60 | — | |
| LLaDA-InstructAttack=PAD, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 61 | — | |
| LLaDA-1.5Attack=DIJA, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 65 | — | |
| LLaDA-InstructAttack=PAD, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 66 | — | |
| LLaDA-InstructAttack=DIJA, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 67 | — | |
| LLaDA-InstructAttack=DIJA, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 67 | — | |
| LLaDA-InstructAttack=DIJA, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 67 | — | |
| LLaDA-InstructAttack=DIJA, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 67 | — | |
| LLaDA-1.5Attack=DIJA, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 68 | — | |
| LLaDA-1.5Attack=DIJA, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 69 | — | |
| LLaDA-InstructAttack=DIJA, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 70 | — | |
| LLaDA-1.5Attack=DIJA, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 70 | — | |
| LLaDA-1.5Attack=DIJA, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 70 | — | |
| LLaDA-InstructAttack=DIJA, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 72 | — | |
| LLaDA-1.5Attack=DIJA, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 72 | — | |
| Dream-InstructAttack=DIJA, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Disabled2026.04 | 73 | — | |
| Dream-InstructAttack=DIJA, Defense=Self-rem., Safety-Aware Denoiser (SAD)=Enabled2026.04 | 73 | — | |
| Dream-InstructAttack=DIJA, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 73 | — | |
| Dream-InstructAttack=DIJA, Defense=None, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 75 | — | |
| Dream-InstructAttack=DIJA, Defense=PPL, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 75 | — | |
| Dream-InstructAttack=DIJA, Defense=PPL, Safety-Aware Denoiser (SAD)=Enabled2026.04 | 75 | — | |
| Dream-InstructAttack=DIJA, Defense=DiffuGuard, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 76 | — | |
| Dream-InstructAttack=DIJA, Defense=None, Safety-Aware Denoiser (SAD)=Disabled2026.04 | 79 | — | |
| Dream-InstructAttack=Zero-shot, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 90 | |
| Dream-InstructAttack=Zero-shot, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 36.2 | |
| Dream-InstructAttack=DIJA, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 0 | |
| Dream-InstructAttack=DIJA, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 0 | |
| Dream-InstructAttack=PAD, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 53 | |
| Dream-InstructAttack=PAD, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 50 | |
| LLaDA-1.5Attack=Zero-shot, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 96 | |
| LLaDA-1.5Attack=Zero-shot, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 8 | |
| LLaDA-1.5Attack=DIJA, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 0 | |
| LLaDA-1.5Attack=DIJA, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 0 | |
| LLaDA-1.5Attack=PAD, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 0 | |
| LLaDA-1.5Attack=PAD, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 1 | |
| LLaDA-InstructAttack=Zero-shot, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 97 | |
| LLaDA-InstructAttack=Zero-shot, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 9 | |
| LLaDA-InstructAttack=DIJA, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 1 | |
| LLaDA-InstructAttack=DIJA, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 0 | |
| LLaDA-InstructAttack=PAD, Safety-Aware Denoiser (SAD)=Disabled2026.04 | — | 0 | |
| LLaDA-InstructAttack=PAD, Safety-Aware Denoiser (SAD)=Enabled2026.04 | — | 4 |