List your company

Jailbreak Defense on Llama-3.1-8B-Instruct

0GCG Success RateQwen3Guard
-3.8422.084873.92Apr 9, 2026

Evaluation results

MethodLinks
Qwen3GuardDefense Method=Qwen3Guard2026.04000131051100035218
TrajGuardDefense Method=TrajGuard (Ours)2026.04010144041031494
Goal PrioritizationDefense Method=Goal Prioritization2026.0417921192094202214541124
Llama Guard 3Defense Method=Llama Guard 32026.04212202522280648612321
Self-GuardDefense Method=Self-Guard2026.04219291720424021018532223
No DefenseDefense Method=No Defense2026.0496834720384769417921687957