Jailbreak Attack on VLGuard Image Unsafe
52.49ASRReAct Dynamic Prompt Rewriting
Evaluation Results
| Method | Links | |
|---|---|---|
| ReAct Dynamic Prompt Rewritingmode=Only ReAct Rewriting2026.01 | 52.49 | |
| ReAct-driven adaptive noising mechanismmode=ReAct & Noise2026.01 | 49.77 | |
| Static Rewritingstrategy=Static template rewriting2026.01 | 36.2 | |
| Originalmode=baseline2026.01 | 33.94 | |
| Only Image Noisingstrategy=Adversarial image noising2026.01 | 32.58 |