Compliance Scoring
The compliance dashboard provides a real-time view of how each agent measures against regulatory requirements. Scores are calculated per agent and per jurisdiction, showing exactly where you stand before enforcement deadlines.
How scores are calculated
Compliance scores are derived from evaluation history across four dimensions:
Violation rate
Total evaluations vs. denied evaluations. A lower violation rate means higher compliance.
Risk tier distribution
How agent actions distribute across minimal, limited, high, and unacceptable risk tiers. Fewer high-risk actions improve the score.
Policy coverage
How many applicable policies have been tested by the agent's evaluation history. Higher coverage means more confidence in the score.
Trend direction
Whether compliance is improving or degrading over time. A downward trend triggers alerts even if the absolute score is acceptable.
Compliance dashboard
The dashboard surfaces compliance data across five views:
- Overall compliance score — percentage per agent, weighted across all active jurisdiction packs
- Per-jurisdiction breakdown — readiness score for each active pack (e.g., EU AI Act 87%, Colorado AI Act 94%)
- Policy violation log — timestamped list of denied actions with the triggering policy and citation
- Risk tier distribution — chart showing how agent actions distribute across risk levels
- Trend lines — compliance trajectory over 7, 30, and 90 day windows
Regulatory readiness
Compliance scoring maps directly to actual regulatory requirements. Each jurisdiction pack defines what “compliant” means for that region:
EU AI Act
Score reflects coverage of transparency, human oversight, and documentation requirements. High-risk system classification drives stricter scoring thresholds. See the EU AI Act guide for full mapping.
Colorado AI Act
Score reflects algorithmic impact assessment and consumer notification readiness. See the Colorado AI Act guide for details.
Related