MEASURE function
The NIST AI RMF MEASURE function covers methods and metrics used to assess AI risks and trustworthiness. Your organization selects the risks, methods, thresholds, and review criteria.
VeriProof can capture supporting records when the integration and governance configuration provide the required signals.
Select measures
Choose measures that answer a defined risk question. Document how each result will inform a decision:
- the risk identified during MAP;
- the data source and calculation;
- the application and population in scope;
- the review frequency and threshold;
- the accountable owner; and
- the action required when the threshold is crossed.
Check source coverage
Not every metric is present in every session. Token counts, latency, model attributes, safety signals, and custom metadata depend on the adapter and application.
Confirm coverage with representative sessions before using a metric in a control or report.
Review results
Use application, session, policy, review, and evidence views available to your role. Compare results with the documented baseline and decision criteria.
Treat missing or partial data as an evidence gap. Do not interpret absence as a successful result.
Verify evidence integrity
Evidence exports include a manifest and proof material for the configured provider. Verification confirms that exported content matches its recorded commitments.
Integrity verification does not measure accuracy, fairness, safety, or regulatory conformity. Those conclusions require separate methods and review.
Record the decision
Record the result, reviewer, rationale, accepted limitations, and follow-up action. Link remediation to the affected application and measure.