Jailbreak Detection on GCG_adapt jailbreak prompts (Llama2-7B) (test)
81ASRHiddenDetect
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HiddenDetectAdaptive loss coefficient (λ)=0.1, Target Model=Llama2-7B2026.06 | 81 | 4 | 9 | 78 | |
| MTKAdaptive loss coefficient (λ)=0.1, Adaptive loss=J3, Target Model=Llama2-7B2026.06 | 72 | 76 | 5 | 17 | |
| HSFAdaptive loss coefficient (λ)=0.1, Target Model=Llama2-7B2026.06 | 67 | 28 | 4 | 48 | |
| SaPAdaptive loss coefficient (λ)=0.1, Target Model=Llama2-7B2026.06 | 62 | 0 | 6 | 62 | |
| GradSafeAdaptive loss coefficient (λ)=0.05, Target Model=Llama2-7B2026.06 | 57 | 12 | 3 | 50 |