Continuous-risk assessment on SIA-Bench SkillJect-derived threat slice (ood)
40.5HR-AUPRCContextual Invocation Audit
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Contextual Invocation Audit2026.04 | 40.5 | 14.6 | |
| Agent-Audit Static Baseline2026.04 | 38 | 11.4 |
Loading the SOTA2 catalog…