The NIST AI Risk Management Framework (AI RMF 1.0), released in January 2023, is a voluntary US framework for managing AI risks. Its core has four functions. Govern is cross-cutting and covers culture, policies and accountability. Map sets context and identifies risks. Measure analyses and tracks them. Manage prioritises and responds. It defines seven characteristics of trustworthy AI and is extended by profiles such as NIST AI 600-1 for generative AI. It's not a certification, but it's widely used and referenced in laws such as Texas TRAIGA.
The NIST AI RMF is one of the most cited AI governance frameworks in the world and a staple of the AIGP exam. Here's how it's built and how organisations use it.
AI RMF at a glance
| Item | Detail |
|---|---|
| Publisher | US National Institute of Standards and Technology (NIST) |
| Released | 26 January 2023 (AI RMF 1.0) |
| Nature | Voluntary, non-sector-specific, use-case agnostic |
| Structure | Part 1 (framing risk, trustworthiness); Part 2 (the Core: four functions, categories, subcategories) |
| Companion resources | AI RMF Playbook; profiles (e.g. NIST AI 600-1 Generative AI Profile, July 2024); crosswalks to other standards |
| Certifiable? | No |
The four functions
| Function | Purpose | Example activities |
|---|---|---|
| Govern | Cultivate a risk-management culture; applies across all other functions | Policies, roles and accountability, workforce diversity and training, third-party risk policies, processes for engaging affected communities |
| Map | Establish context and identify risks | Define intended purpose and context, categorise the system, map impacts on individuals and society, identify third-party components |
| Measure | Analyse, assess and track risks | Choose metrics, test for trustworthiness characteristics, evaluate bias, track risks over time, gather feedback |
| Manage | Prioritise and act on risks | Risk treatment, response and recovery plans, monitoring after deployment, decommissioning |
Govern underpins the others. Map, Measure and Manage aren't strictly sequential and are applied iteratively through the AI lifecycle.
Seven characteristics of trustworthy AI
- Valid and reliable
- Safe
- Secure and resilient
- Accountable and transparent
- Explainable and interpretable
- Privacy-enhanced
- Fair, with harmful bias managed
NIST treats validity and reliability as a base for the others, and accountability and transparency as relating to all of them. Trade-offs between characteristics, such as accuracy against interpretability, are expected and should be managed explicitly.
Profiles: the Generative AI Profile (NIST AI 600-1)
Profiles apply the framework to specific settings. NIST AI 600-1 (July 2024) identifies risks that are unique to or made worse by generative AI. These include confabulation, harmful content, data privacy, information integrity, information security, intellectual property, harmful bias and homogenisation, and value-chain and component integration. It also suggests actions mapped to the AI RMF functions. Texas TRAIGA treats substantial compliance with this profile, or similar frameworks, as a defence. See Texas TRAIGA.
Related NIST work
- NIST AI 100-2: a taxonomy of adversarial machine learning attacks and mitigations, useful for red teaming. See GenAI red teaming for the AIGP.
- CAISI: NIST's Center for AI Standards and Innovation (formerly the US AI Safety Institute), which focuses on evaluating frontier models.
- Updates: the 2025 US AI Action Plan directed NIST to revise the AI RMF. Check NIST's site for the current version before relying on specific wording.
The categories inside each function
Each function is broken into categories and subcategories. You don't need every subcategory for the AIGP, but knowing what each category is about helps with "which function?" questions.
| Function | Categories cover |
|---|---|
| Govern (6 categories) | Policies and processes; accountability structures; workforce diversity, equity and inclusion; a culture of critical thinking and safety; engagement with external stakeholders; third-party and supply chain risks |
| Map (5 categories) | Context and intended purpose; categorising the system; capabilities, goals and expected benefits and costs; risks and benefits of third-party components; impacts on individuals, groups, communities and society |
| Measure (4 categories) | Selecting methods and metrics; evaluating trustworthiness characteristics; tracking risks over time; gathering feedback on measurement effectiveness |
| Manage (4 categories) | Prioritising and responding to risks; strategies to maximise benefits and minimise harms; managing third-party risks; documenting and monitoring risk treatments, including response and recovery |
The AI RMF Playbook and profiles
- The Playbook gives suggested actions, references and documentation ideas for each subcategory. It's voluntary and meant to be adapted, not followed line by line.
- Profiles tailor the framework to a sector, technology or use case. A "current profile" describes where you are now; a "target profile" describes where you want to be, and the gap between them becomes a roadmap.
- NIST AI 600-1 (the Generative AI Profile, July 2024) identifies risks that are unique to or made worse by generative AI, such as confabulation, harmful content, information integrity, data privacy, intellectual property and value chain issues, and maps suggested actions to the four functions.
A worked example: applying the AI RMF to a customer chatbot
| Function | Example actions |
|---|---|
| Govern | Assign an accountable owner; set a policy on what the bot may and may not answer; define risk tolerance for wrong answers |
| Map | Define intended use (order questions only); identify users, including vulnerable customers; list harms such as wrong refund information or data leakage |
| Measure | Test answer accuracy on real queries; red team for prompt injection and data disclosure; track hallucination rates |
| Manage | Add guardrails and escalation to a human; monitor complaints; plan rollback; review after incidents |
A real case shows why this matters: in Moffatt v Air Canada (2024), a Canadian tribunal held the airline responsible for incorrect refund information its chatbot gave a customer.
NIST AI RMF vs ISO/IEC 42001 vs EU AI Act
| NIST AI RMF | ISO/IEC 42001 | EU AI Act | |
|---|---|---|---|
| Type | Voluntary framework | Certifiable standard | Binding law |
| Focus | Risk practices | Management system | Obligations per system and role |
| Proof | Self-attestation | Certification audit | Conformity assessment |
See ISO/IEC 42001 explained and Singapore's framework vs NIST.
How to use the AI RMF in practice
- Start with Govern: an AI policy, risk appetite, roles and an inventory.
- For each system, Map its context and impacts.
- Measure against the trustworthiness characteristics with defined metrics.
- Manage: decide to treat, accept, transfer or avoid each risk, and monitor.
- Use the Playbook for suggested actions, and build a profile for your sector or use case.
Frameworks you'll be tested on
The AIGP Complete Pack covers the NIST AI RMF, the GenAI profile, ISO/IEC 42001 and 42005 in its Cram Guide and BoK v2.1 practice questions.
Get the Complete PackFrequently asked questions
What are the four functions of the NIST AI RMF?
Govern, Map, Measure and Manage. Govern is cross-cutting and supports the other three.
Is the NIST AI RMF mandatory?
No. It's voluntary, though some laws and contracts reference it, and US federal agencies have used it in their own AI policies.
Is there a NIST AI RMF certification?
No. There's no organisational certification or individual credential for the AI RMF.
What is NIST AI 600-1?
The Generative AI Profile of the AI RMF, published in July 2024. It identifies generative AI risks and suggests actions for managing them.
What are the characteristics of trustworthy AI in the NIST AI RMF?
Valid and reliable; safe; secure and resilient; accountable and transparent; explainable and interpretable; privacy-enhanced; and fair, with harmful bias managed.
What is the NIST AI RMF Playbook?
A companion resource with suggested actions, references and documentation ideas for each subcategory of the AI RMF. It is voluntary and meant to be adapted to each organisation.
What is a NIST AI RMF profile?
A tailored version of the framework for a sector, technology or use case. Comparing a current profile with a target profile helps an organisation plan improvements. NIST AI 600-1 is the profile for generative AI.
Sources
- NIST AI Risk Management Framework
- NIST AI 100-1: AI RMF 1.0 (PDF)
- NIST AI 600-1: Generative AI Profile (PDF)
- NIST Trustworthy and Responsible AI Resource Center
Last reviewed: October 2026.