Before a user test, define the system’s intended task and deployment context, identify who may be affected, set data boundaries, and decide how performance and failure will be evaluated. NIST’s voluntary AI RMF organizes risk work under Govern, Map, Measure and Manage, offering a practical structure for these questions.
Direct answer
Before a user test, define the system’s intended task and deployment context, identify who may be affected, set data boundaries, and decide how performance and failure will be evaluated. NIST’s voluntary AI RMF organizes risk work under Govern, Map, Measure and Manage, offering a practical structure for these questions.
A practical sequence
- Govern: name accountable owners and review responsibilities.
- Map: document task, user, affected groups, data, workflow and plausible harms.
- Measure: evaluate representative cases, edge cases, uncertainty and baseline performance.
- Manage: set human review, escalation, fallback, monitoring and stop conditions.
- Document consent, data handling, limitations and the decision to proceed or pause.
Limits and boundaries
The NIST framework is not a certification or substitute for legal, security, privacy, clinical or sector-specific review. The appropriate checks depend on use and jurisdiction.
Sources and review
- NIST AI RMF — Core functionsAuthoritative context for this original educational answer.
General education, not individualized business, investment, legal, technical or regulatory advice. Reviewed 31 October 2026. Context and requirements vary; consult qualified specialists where appropriate.