Jailbreak Detection on Llama-3-8B (JBShield GCG)
100Detection AccuracyRTV
Evaluation Results
| Method | Links | |
|---|---|---|
| RTVTarget Model=Llama-3-8B, n (Sample Size)=164, FPR Threshold=5%2026.05 | 100 | |
| JBShield-DTarget Model=Llama-3-8B, n (Sample Size)=50, FPR Threshold=5%2026.05 | 98 | |
| RTVTarget Model=Llama-3-8B, n (Sample Size)=50, FPR Threshold=5%2026.05 | 88 | |
| JBShield-DTarget Model=Llama-3-8B, n (Sample Size)=164, FPR Threshold=5%2026.05 | 0 |