Prompt Injection Detection on Indirect-Hijack Shift
69.3False Negative Rate (FNR)ProtectAI-v2
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ProtectAI-v2t=0.0282026.06 | 69.3 | 99.3 | 104 | 39.7 | 0.424 | |
| Prompt-Guard-2 (86M)t=0.0042026.06 | 69.3 | 99.8 | 104 | 14.5 | 0.625 |
Loading the SOTA2 catalog…