List your company

Jailbreak Attack on VLGuard Image Unsafe

52.49ASRReAct Dynamic Prompt Rewriting
31.783637.159342.53547.9107Jan 29, 2026

Evaluation results

MethodLinks
ReAct Dynamic Prompt Rewritingmode=Only ReAct Rewriting2026.0152.49
ReAct-driven adaptive noising mechanismmode=ReAct & Noise2026.0149.77
Static Rewritingstrategy=Static template rewriting2026.0136.2
Originalmode=baseline2026.0133.94
Only Image Noisingstrategy=Adversarial image noising2026.0132.58