List your company

Jailbreak Defense on Mistral-7B Evaluation

0GCG Attack Success RateQwen3Guard
-3.9622.7749.576.23Apr 9, 2026

Evaluation results

MethodLinks
Qwen3GuardDefense Method=Qwen3Guard2026.04000251051300035219
Llama Guard 3Defense Method=Llama Guard 32026.042124131253307011652726
TrajGuardDefense Method=TrajGuard (Ours)2026.0431261916471142191410
Goal PrioritizationDefense Method=Goal Prioritization2026.044718440315767124232641944
Self-GuardDefense Method=Self-Guard2026.044701382714288170373121
No DefenseDefense Method=No Defense2026.0499969741467875698535918275