List your company

Skill Invocation Safety Auditing on SIA-Bench Locked in-distribution (test)

20.4HR-AUPRCFusion
16.7617.70518.6519.595Apr 11, 2026

Evaluation results

MethodLinks
Fusion2026.0420.411.8200.09827.732.1
Contextual2026.0419.510.517.80.11221.628.8
Text-Only2026.0418.96.611.10.07321.728.8
Static Prior2026.0418.812.220.60.03114.929.1
Denylist2026.0418.812.220.60.03114.929.1
No Audit2026.0416.91016.9032.132.1
Agent-Audit2026.0416.9915.20.01531.831.9