Jailbreak Attack on VLGuard Text Unsafe
34.59ASRReAct-driven adaptive noising mechanism
Evaluation Results
| Method | Links | |
|---|---|---|
| ReAct-driven adaptive noising mechanismmode=ReAct & Noise2026.01 | 34.59 | |
| ReAct Dynamic Prompt Rewritingmode=Only ReAct Rewriting2026.01 | 32.8 | |
| Static Rewritingstrategy=Static template rewriting2026.01 | 15.41 | |
| Originalmode=baseline2026.01 | 6.81 | |
| Only Image Noisingstrategy=Adversarial image noising2026.01 | 6.26 |