The NIST AI Risk Management Framework (AI RMF 1.0) is a free, voluntary framework from the US National Institute of Standards and Technology, published in January 2023. It’s become one of the most widely used references for managing AI risk, inside and outside the US, because it’s practical, free and doesn’t require certification.
Who it’s for
Any organisation that designs, develops, buys or uses AI, of any size and in any sector. It isn’t a law, so nobody is forced to use it. Companies adopt it because it gives them a structured, recognised way to show they manage AI risk responsibly. That helps with customers, auditors and regulators, and maps well onto the EU AI Act.
What “trustworthy AI” means in the framework
NIST describes seven characteristics of trustworthy AI:
- Valid and reliable: it works as intended.
- Safe: it doesn’t endanger people.
- Secure and resilient: it withstands attacks and failures.
- Accountable and transparent: someone answers for it, and information about it is available.
- Explainable and interpretable: people can understand how and why it produces outputs.
- Privacy-enhanced: it protects personal data.
- Fair, with harmful bias managed.
These often pull against each other. More explainability can cost accuracy, for example. Part of the job is deciding the right balance for each use.
The four core functions
The framework’s “Core” is built around four functions.
Govern
Build the culture and structure for AI risk management: policies, roles, accountability, training, and processes for third-party AI. Govern applies across the other three and is the foundation.
Map
Understand the context of each AI system: what it’s for, who it affects, what could go wrong, and what assumptions it relies on. You can’t measure a risk you haven’t identified.
Measure
Analyse and track the risks you mapped, using tests, metrics and evaluations, for example accuracy, bias, robustness and security testing, before and after deployment.
Manage
Prioritise and act on the risks: decide what to fix, accept, transfer or avoid, respond to incidents, and keep monitoring over time.
Each function breaks down into categories and subcategories, which work like a checklist of outcomes.
Companion resources
- The AI RMF Playbook suggests actions for each subcategory.
- The Generative AI Profile (NIST AI 600-1), published in July 2024, applies the framework to generative AI risks such as confabulation, data privacy, information security and harmful content.
How to use the NIST AI RMF in practice
- Start with Govern. Name owners, write a short AI policy, and decide who approves new AI uses.
- Map your systems. Use your AI inventory and write a short context statement for each important system.
- Measure what matters. Pick a few meaningful tests per system, for example error rate on real cases or bias across groups, rather than measuring everything.
- Manage and repeat. Track risks in a register, assign actions, and review at least every quarter.
NIST AI RMF vs ISO/IEC 42001 vs the EU AI Act
- NIST AI RMF: a voluntary, flexible framework with no certification. Good for building the practice.
- ISO/IEC 42001: a certifiable management-system standard. Good when customers want formal proof.
- EU AI Act: the law, with mandatory obligations for organisations in scope.
Many organisations use NIST AI RMF to structure the work, ISO/IEC 42001 to certify it, and the EU AI Act to set the minimum legal bar.
Official sources
- NIST AI Risk Management Framework: the framework and its supporting documents
- NIST AI 600-1, Generative AI Profile: the AI RMF applied to generative AI
- NIST AI Resource Center: the AI RMF Playbook and crosswalks to other frameworks
More primary sources, with a short note on each, are on our Sources page.
Last reviewed October 2026. Based on NIST AI 100-1 (AI RMF 1.0) and NIST AI 600-1. This article is general information.