The NIST AI Risk Management Framework organizes AI risk work into four functions. Govern sets policies and accountability. Map frames the context of a system. Measure tests and tracks risk. Manage prioritizes and responds. Govern applies to everything else, while Map, Measure and Manage run in a loop for each AI system.
Govern: the function that touches everything
Govern is about the organization, not any single model. NIST describes it as cross-cutting. Its categories cover policies and processes, accountability structures, workforce, a culture that communicates AI risk, engagement with outside parties, and third-party and supply chain risk.
Practical outputs include an AI policy, named owners, training, and an inventory of AI systems. The framework lists the inventory under GOVERN 1.6.
Map: establish context before you judge risk
Map asks what the system is for, who uses it, what laws and norms apply, and what could go wrong. It also covers categorizing the system (a classifier, a generative model, a recommender), documenting knowledge limits, and mapping risks from third-party components.
The framework says that after Map, you should have enough context to make an initial go or no-go decision on whether to build or deploy the system.
Measure: test, then keep testing
Measure uses quantitative, qualitative or mixed methods to analyze and monitor AI risk. Systems should be tested before deployment and regularly while in operation. The subcategories cover documenting test sets and metrics, evaluating safety, security and resilience, privacy, fairness and bias, and whether the measurement methods themselves work.
Manage: decide, respond, recover
Manage turns Map and Measure output into action. Risks are prioritized, responses are planned and documented, and the options include mitigating, transferring, avoiding or accepting. It also covers third-party risk, post-deployment monitoring, and incident communication. MANAGE 2.4 asks for mechanisms to supersede, disengage or deactivate a system that behaves outside its intended use.
How the four fit together
Think of Govern as the rules of the road and the other three as a cycle. Map frames, Measure checks, Manage acts, and new findings feed back into Map as context changes. NIST notes it is on the user to keep applying Map and Measure as systems and risks evolve.
You do not need to implement every subcategory on day one. Start with the inventory, an owner per system, and a written risk tolerance, then add testing depth where the risk is highest.
A concrete example
Take a support chatbot that answers billing questions. Govern gives it an owner, a policy and a spot in the inventory. Map records that it handles customer account data, is wrong sometimes, and must hand off to a person for refunds. Measure runs a test set of real billing questions and watches production complaints. Manage decides what happens when accuracy drops: tighten the prompt, limit the data it can see, or switch it off. The loop then repeats with what you learned.
Quick check
Which function comes first?
Govern is foundational and runs through the others. In practice you begin with enough Govern (policy, owners, inventory) to Map your first system.
Is this a certification?
No. The AI RMF is a voluntary framework, not a certification scheme.
Sources
Last reviewed: 19 October 2026.
