Jailbreak Attack on VLGuard Safe
8.44Attack Success Rate (ASR)Only Image Noising
Evaluation Results
| Method | Links | |
|---|---|---|
| Only Image Noisingstrategy=Adversarial image noising2026.01 | 8.44 | |
| ReAct-driven adaptive noising mechanismmode=ReAct & Noise2026.01 | 7.53 | |
| Originalmode=baseline2026.01 | 7.35 | |
| ReAct Dynamic Prompt Rewritingmode=Only ReAct Rewriting2026.01 | 7.17 | |
| Static Rewritingstrategy=Static template rewriting2026.01 | 5.56 |