AI & Business · Sourced explainer

NIST AI Governance: Four Functions in Practice

Understand NIST's Govern, Map, Measure, and Manage functions, then translate them into a practical review record for a business AI use case.

Illustration of a computer model linked to a clipboard, map, ruler, and human decision gate
AI-generated editorial illustration. Not a photograph of an event.

NIST's AI Risk Management Framework provides a vocabulary for identifying and managing AI risks. Its four core functions are Govern, Map, Measure, and Manage. They connect oversight, context, evaluation, and action rather than forming a simple linear checklist.

The framework is voluntary. NIST's current pages say AI RMF 1.0 is being revised, so this explainer describes the published 1.0 structure rather than claiming that it is a newly issued regulation or certification. NIST framework overview

Govern: assign decisions to people

NIST places governance across the other functions. A practical business record can name the process owner, the person authorized to approve deployment, and the route for reporting a problem. AI RMF Core

Consider a hypothetical assistant that drafts customer replies. Someone must decide which subjects it may answer, whether a draft needs review, and who can suspend the feature. A vendor contract does not answer those operational questions by itself.

Map: describe the use and the affected people

Document what the system does, the information it receives, and where its output goes. A draft saved for an employee differs materially from an answer automatically sent to a customer.

For the hypothetical reply assistant, map the support categories, customer data, language coverage, and possible consequences of an incorrect response. Include workarounds: an employee might paste confidential material into a tool that was approved only for public documents. The actual workflow matters more than the intended diagram.

Measure: choose evidence that matches the consequence

A proposed evaluation could use a set of approved test questions with known answers, edge cases, and requests the assistant should escalate. Record correctness, omitted context, inappropriate certainty, and the time reviewers spend repairing drafts.

Do not reduce every outcome to a single accuracy score. A mistaken opening-hours answer and a fabricated refund promise may have different consequences. State which errors are release-blocking and how that decision was made. The test set should represent the intended use and be revisited when that use changes.

Manage: decide what happens next

Use the evidence to decide whether to launch, restrict, change, or stop the use case. Write down the unresolved risks, the person accepting them, and the conditions that trigger another review.

For the same example, the company might allow draft responses to basic product questions while excluding payment disputes. That is an illustrative deployment choice, not a rule imposed by NIST. A useful record connects a limitation to an actual control, such as routing excluded categories to a person.

Keep a compact use-case record

A practical template can contain purpose, owner, users, data inputs, output destination, vendor version, test evidence, unresolved issues, and review date. Link to the detailed evidence instead of copying it into several documents.

Review after a material model change, new integration, serious incident, or expansion to a different customer group. A previously acceptable system can become unsuitable when its authority or context changes.

The AI workflow guide addresses operating value, while the cybersecurity guide covers account access and recovery. Governance should connect those questions to the decision about acceptable use.

Common questions

Does using the AI RMF prove legal compliance?

No. A voluntary risk framework does not replace obligations that apply to a particular industry, jurisdiction, or use case.

Are the four functions a one-time approval sequence?

NIST describes related risk-management activities rather than a fixed checklist. Revisit them as evidence and use change.

Can a small company use the framework without a dedicated committee?

A company can scale its documentation and ownership to the task. The essential point is that relevant decisions and evidence have accountable owners.

References

Sources & further reading

Sources consulted October 17, 2026. These organizations do not endorse this article.

  1. AI Risk Management FrameworkNational Institute of Standards and Technology
  2. AI RMF CoreNIST AI Resource Center

Explore the numbers

Calculators use your inputs and assumptions. They do not validate a business forecast.

LLM Token EstimatorEstimate LLM API costs based on token usageSecurity Incident Cost CalculatorCalculate total cost of security incidents

Continue reading

How to Choose a Small-Business AI WorkflowAI & Business · 4 min readCybersecurity Basics for a Growing BusinessCybersecurity · 4 min read
← All articles