The NIST AI Risk Management Framework is a voluntary framework for organisations that design, develop, deploy or use AI. Its four functions—Govern, Map, Measure and Manage—are not a product checklist. They describe connected risk-management work that should continue throughout the AI lifecycle.
For buyers, that distinction matters. A supplier may support part of a function without supporting the whole operating process. The aim is not to find a platform that “complies with NIST” in the abstract, but to understand which parts of the organisation’s governance model the platform can enable and evidence.
Govern: establish authority and accountability
Govern is cross-cutting. It includes policies, roles, responsibilities, risk tolerances and oversight. In product terms, buyers should look for ownership records, approval workflows, policy and control mapping, lifecycle states, exception handling and an auditable decision history.
A dashboard alone does not establish governance. Ask how responsibilities are assigned, how an orphaned system is escalated, how material changes trigger reassessment and how accepted risks are recorded.
Map: establish context before scoring risk
Map is about understanding the intended purpose, users, environment, dependencies and potential impacts of an AI system. Risk cannot be interpreted without this context. The same model may create very different consequences when used for drafting marketing copy and when used to influence employment or access decisions.
Relevant capabilities include structured intake, system and agent inventories, data and tool relationships, stakeholder identification, use-case classification and dependency mapping. Automated discovery can add visibility, but discovered assets still need business context and ownership.
Ask the supplier to show how one discovered AI asset becomes a contextualised governance record with purpose, owner, dependencies and lifecycle status.
Measure: evaluate risk with evidence
Measure covers analysis, assessment, testing and monitoring. A platform may contribute by organising assessments, collecting evidence, tracking metrics, monitoring behaviour or integrating specialist testing tools.
Buyers should check how a risk rating was produced, which evidence supports it, when the evidence was collected and whether changes in the system can alter the assessment. For generative and agentic AI, monitoring may also need to cover prompts, responses, tool calls, permissions, anomalous behaviour and human escalation.
Manage: prioritise and act
Manage converts findings into treatment decisions. This may include accepting, reducing, transferring or avoiding risk; assigning remediation; restricting access; adding approval gates; or suspending a system. Strong products connect the finding to a decision, authorised action, responsible owner and verification step.
The framework does not require that one platform perform every action itself. Procurement should therefore identify where the product is the system of record, where it orchestrates another control and where it only provides a recommendation.
What the Generative AI Profile adds
NIST’s Generative AI Profile applies the AI RMF to risks that are especially relevant to generative AI. It gives organisations additional considerations and actions rather than replacing the four-function model. For buyers, it reinforces the need to assess provenance, content risks, information security, human oversight, evaluation and monitoring in the intended context.
Procurement questions derived from the RMF
- How does the product assign and maintain accountable ownership?
- Can it connect an AI system to data, identities, tools, APIs and downstream processes?
- What evidence supports each risk score or control status?
- How does a material change trigger review or reassessment?
- Which controls can the product enforce, and which require integrations?
- How does it prove that remediation changed the risk or system state?
These questions keep the evaluation focused on operating capability. They also make supplier comparisons more defensible because each answer can be linked to a governance outcome and evidence.