Jailbreak Detection on JailbreakBench AutoDAN-T
76Detection RateGuardReasoner
Evaluation Results
| Method | Links | |
|---|---|---|
| GuardReasonerTarget Model=Vicuna2026.05 | 76 | |
| GuardReasonerTarget Model=Llama22026.05 | 68 | |
| embedding disruption methodTarget Model=Llama22026.05 | 68 | |
| GuardReasonerTarget Model=Qwen2.52026.05 | 65 | |
| embedding disruption methodTarget Model=Vicuna2026.05 | 65 | |
| embedding disruption methodTarget Model=Qwen2.52026.05 | 52 | |
| Llama-Guard-3Target Model=Vicuna2026.05 | 49 | |
| Llama-Guard-3Target Model=Llama22026.05 | 32 | |
| Llama-Guard-3Target Model=Qwen2.52026.05 | 18 |