Jailbreak Detection on AdvBench
97.5Refusal RateLWV
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LWVEvaluation Protocol=Zero-shot2026.06 | 97.5 | 2.5 | |
| MLEvaluation Protocol=Zero-shot2026.06 | 97.31 | 2.69 | |
| NeMoGuardEvaluation Protocol=Zero-shot2026.06 | 85.58 | 14.42 | |
| ProtectAI v1Evaluation Protocol=Zero-shot2026.06 | 0.58 | 99.42 |