START HERE

What you will learn

  • Explain Govern, Map, Measure and Manage in a concrete use case.
  • Distinguish application security checks from broader AI risk decisions.
  • Create evidence for a limited release decision and later review.

Before you start: The workflow-evaluation guide is a useful starting point. This article explains the published 1.0 structure; check NIST for revisions before adopting it.

Reference edition: NIST AI RMF 1.0; official revision in progress at the source-check date. Sources checked 2026-11-11.

Security is part of the picture

Imagine an assistant that only retrieves documents a person may access. The access rule might work correctly while the summary omits a crucial qualification and leads the reader towards a poor recommendation. That failure needs attention even if no security boundary was crossed.

NIST’s published AI RMF 1.0 organises work under Govern, Map, Measure and Manage. Governance runs across the work; the functions are not necessarily an ordered checklist. NIST states that this edition is being revised, so this guide names the version it is explaining rather than treating it as an unchanging final answer.

Connect the functions to decisions you can explain

Govern concerns responsibility and the rules for use. Map examines the purpose, context, affected people and potential consequences. Measure gathers suitable evidence about relevant risks and behaviour. Manage uses that evidence to choose actions, limits and follow-up.

For a technical leader, the useful result is not four headings in a policy. It is a record of what the system is for, what could go wrong, how you assessed it and who can restrict or stop its use. Include the people affected by the output where their perspective changes the assumptions.

Step-by-step process

  1. Define the intended purpose and accountable decision

    State the task, intended users and decisions the system may support. Name who can authorise the scope and respond to a serious concern.

  2. Describe the context and possible effects

    List the relevant information, affected people, dependencies and ways an output could be misused or misunderstood. Consider whether AI is appropriate for the task at all.

  3. Choose evidence that matches the risk

    Use test cases, source checks, error analysis and other appropriate methods. A generic accuracy score may not reveal omitted qualifications or reliance on unsupported claims.

  4. Decide controls and release limits

    You might require source verification, exclude particular tasks or pause deployment until a failure is resolved. Explain why the measures address the observed risk.

  5. Monitor and revisit the decision

    Track meaningful failures and changes in context. Define when the use must be reassessed, who handles reports and how restrictions or shutdown would be applied.

FOLLOW ONE EXAMPLE

Worked scenario

A stakeholder assistant that writes confident recommendations

A fictional assistant summarises supplier options. It respects document permissions but repeatedly omits exclusions that make an option unsuitable for the stakeholder’s intended use.

  1. The owner maps how readers use the brief and identifies omission of a decisive qualification as an important failure.
  2. Evaluation includes documents with exclusions and requires the brief to make them visible. The result is assessed alongside source references, not just readability.
  3. The release is limited to checked drafts while the failure is investigated. The owner records the restriction, responsible reviewer and evidence needed before expanding use.

What this shows: The release decision reflects the consequences of the output. Passing an access-control test alone would not resolve the recommendation problem.

The process at a glance

Governance informs the other functions, while evidence and decisions feed back into the use-case assessment. The arrows illustrate a working relationship, not a prescribed linear checklist.
ProcessDecisionHuman checkpoint
Read Mermaid diagram source
flowchart TD
 G[Govern - responsibilities and rules] --> M[Map - purpose and effects]
 G --> E[Measure - collect relevant evidence]
 G --> A[Manage - limits and actions]
 M --> E
 E --> A
 A --> M
 A --> G

What you should have at the end

  • A clear intended-use statement and risk owner.
  • Evaluation cases connected to the effects that matter in the use case.
  • Recorded restrictions, monitoring and reassessment decisions.

Common mistakes and how to avoid them

Calling security testing the whole AI risk assessment

Also examine output suitability, context and potential effects on people.

Using one score without inspecting failure types

A readable answer can still omit the fact that changes the decision.

Treating the framework functions as a one-time sequence

Keep governance and reassessment active throughout use.

APPLY THE IDEA

Try it yourself

Your task: Choose a proposed AI use case. Write one intended benefit, one affected person or group and one plausible harmful outcome.

What to produce: A short note connecting that outcome to an evaluation case, a proposed limit and the owner of the release decision.

Check your understanding: Would your test reveal the harmful outcome, or only whether the output looks fluent?

Key terms in plain English

Intended use
The task and context the system is designed and permitted to support.
AI risk
The possibility and effects of unwanted outcomes associated with an AI system.
Risk treatment
A decision or measure intended to address an identified risk.
Reassessment trigger
A change or observation that requires the previous decision to be reviewed.

Sources and further reading