Benchmark RadarAI BENCHMARK PROFILE
Fiducia-bench
Evaluates governability of financial agents across 626 episodes and 100 KYC/AML task variants, measuring escalation, abstention, and audit trail completeness under different agent architectures.
- Released
- 2026-08-17
- Readiness
- Paper only
- Primary field
- Finance & Economics
Why it matters
Identifies a governance gap in decomposed agent architectures, providing a framework to assess whether policy compliance is preserved when agents are broken into components.
Motivation
Existing agent benchmarks ask whether the agent finished the task.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.