Jailbreak Detection on Red-teaming Queries high-risk vulnerabilities
1.871High-Risk Vulnerability Discovery RateLast Generation
Evaluation Results
| Method | Links | |
|---|---|---|
| Last GenerationRelative Compute Overhead (%)=100.02026.01 | 1.871 | |
| Exchange ClassifierRelative Compute Overhead (%)=150.02026.01 | 0.885 | |
| Two-Stage CascadeRelative Compute Overhead (%)=27.82026.01 | 0.878 | |
| Production GradeRelative Compute Overhead (%)=3.52026.01 | 0.505 |