The NIST AI Risk Management Framework is a voluntary framework for managing risks AI creates for people, organisations and society. Its core has four functions: Govern, Map, Measure and Manage. This glossary explains the main terms in plain words, so you can read the framework and answer customer questions accurately.
What is the AI RMF itself?
NIST released AI RMF 1.0 on 26 January 2023 after a public process that included requests for information, drafts and workshops. NIST says it is intended for voluntary use. There is no certificate and no audit built into it. NIST's page also says the framework is being revised, so check the current version.
The four functions
Govern. Building a culture and structure for managing AI risk: policies, processes, accountability, and ties to organisational values. NIST describes it as cross-cutting, meaning it informs the other three. It covers the whole lifecycle, including third-party software, hardware and data.
Map. Setting the context. What is the system for, who uses it, where is it deployed, and what could go wrong for individuals, groups and society? The output supports an early decision about whether to proceed at all.
Measure. Assessing and tracking risk with quantitative, qualitative or mixed methods. It includes testing before deployment and monitoring while the system runs, against characteristics such as safety, security, fairness, privacy and transparency.
Manage. Deciding what to do about the risks you found. Responses include mitigating, transferring, avoiding or accepting. It also covers incident response, recovery, communication, third-party monitoring and switching off a system that behaves outside its intended use.
NIST says the core is not a checklist, and you can apply functions in any order that adds value, iterating as you learn.
Other terms you will meet
Core. The set of outcomes the framework describes, organised by function and sub-category.
Sub-category. The detailed level below a function. The Playbook aligns its suggestions to each sub-category.
Playbook. A companion resource of suggested actions for reaching the core's outcomes. NIST calls it neither a checklist nor a set of steps to follow in full. Use as much of it as fits your case.
Profile. A tailored application of the core to your setting, requirements, risk tolerance and resources. NIST does not prescribe a template. Profiles come in a few forms:
- A use-case profile applies the framework to a specific application or sector.
- A temporal profile has a current version (how you manage AI today) and a target version (where you want to be). The gap between them becomes your action plan.
- A cross-sectoral profile covers risks common to many uses, such as large language models or cloud services.
Generative AI Profile. NIST-AI-600-1, released on 26 July 2024. It helps organisations identify risks specific to generative AI and suggests actions aligned to their goals.
Trustworthy AI characteristics. NIST lists seven: valid and reliable; safe; secure and resilient; accountable and transparent; explainable and interpretable; privacy-enhanced; and fair, with harmful bias managed. Valid and reliable is treated as the base for the others. Where tradeoffs arise, NIST says to resolve them in a justifiable way and notes that thresholds need human judgment.
What does this mean for a customer questionnaire?
If a customer asks whether you follow the AI RMF, the honest answers are specific. Say which functions you have processes for, who owns them, and what evidence exists. Do not say you are certified, because the framework has no certification. A target profile and a gap list is a good thing to show.
Tools can organise the evidence. Safeguard's compliance automation tracks frameworks, controls and evidence in one place, and Safeguard can run with Full AI, Zero AI or your own model, which is a deployment fact you may need to state. The framework is guidance you apply yourself.
What to do
- Read the core and the Generative AI Profile once, start to finish.
- List every AI system you build or buy, and name an owner for each.
- For each, write the purpose, users and deployment context. That is Map.
- Choose which trustworthy characteristics matter most for each use.
- Write a current profile, then a target profile, and list the gaps.
- Define how you test, monitor and switch off each system.
- Review the list on a schedule, and record changes.
Quick check
Is there a NIST AI RMF certification?
No. NIST describes the framework as voluntary, and the sources we read describe no certification. You can describe your alignment, but not claim certification.
Do I have to complete every Playbook action?
No. NIST says the Playbook is not a checklist, and organisations can adopt as many or as few suggestions as fit their use case.
Sources
- NIST: AI Risk Management Framework
- NIST AIRC: AI RMF Playbook
- NIST AIRC: AI RMF Core
- NIST AIRC: Characteristics of trustworthy AI
- NIST AIRC: AI RMF Profiles
Last reviewed: 18 October 2026.
