The NIST AI Risk Management Framework (AI RMF 1.0) is a voluntary guide for managing AI risk, built around four functions: Govern, Map, Measure and Manage. It is not a certification. A practical start is to list your AI systems, name an owner for each, and work through the functions per system.
What is it and who is it for?
NIST published AI RMF 1.0 on 26 January 2023 as NIST AI 100-1, developed through an open process of requests for information, draft versions, workshops and public comment. NIST describes it as voluntary, rights-preserving, non-sector-specific and use-case agnostic, meant to help organizations that design, develop, deploy or use AI incorporate trustworthiness into their work.
Two things about its status are worth knowing. NIST's page says the AI RMF 1.0 is being revised as part of the White House AI Action Plan, so check the NIST page for the current version before you cite it. And NIST has published companion material: a Playbook, a Generative AI Profile released on 26 July 2024 (NIST AI 600-1), and, as of 7 April 2026, a concept note for a profile on trustworthy AI in critical infrastructure.
What does "trustworthy" mean here?
Part 1 of the framework lists seven characteristics of trustworthy AI systems: valid and reliable; safe; secure and resilient; accountable and transparent; explainable and interpretable; privacy-enhanced; and fair, with harmful bias managed. The framework acknowledges these involve trade-offs. A system can be accurate but not secure, or secure but opaque, and the point is to decide those trade-offs on purpose.
What are the four functions?
The Core divides work into functions, categories and subcategories.
Govern applies across everything. It covers legal and regulatory requirements, policies and processes, roles and accountability, and decisions about risk tolerance. NIST notes that Govern applies to all stages of the organization's AI risk management, while the other three can be applied to a specific system at a specific lifecycle stage.
Map establishes context. Its subcategories cover intended purposes and settings of use, who the users are, benefits, costs and risks, and the likelihood and magnitude of impacts.
Measure is where you test and track. It covers choosing metrics and methods, evaluating systems for trustworthy characteristics, and monitoring over time.
Manage is the response. Subcategories include deciding whether a system should proceed, prioritizing risk treatment by impact, likelihood and resources, handling risks from third-party components, and planning post-deployment monitoring and incident response. Risk response options include mitigating, transferring, avoiding or accepting.
A constructed example
A company adds an AI summarizer to its customer support tool. Under Govern, the CTO names an owner and the legal lead confirms the data handling rules. Under Map, the owner writes down the purpose (summarize tickets for agents), the users (support staff), and the risks (summaries that omit a safety complaint, customer data leaving the tool). Under Measure, the team builds a test set of past tickets with known key facts and tracks how often summaries miss them. Under Manage, they decide that summaries can be shown to agents but not sent to customers, set a threshold for pausing the feature, and review third-party model terms. The scenario is invented, but each step is a small, ordinary piece of the framework.
How to check or apply this
- Build an inventory of AI systems, including those bought as features inside other products.
- Name an accountable owner for each system.
- Write a one-page Map for each: purpose, users, data, failure modes, affected people.
- Pick two or three measurable tests per system and run them before launch and on a schedule after.
- Decide who can pause or retire a system and what triggers it.
- Review third-party models and data suppliers as part of Manage.
- Keep the evidence: decisions, test results and incident notes, in one place.
- Read the Generative AI Profile if you deploy generative systems.
Quick check
Can a company be certified against the NIST AI RMF?
No. It is a voluntary framework, not a certification scheme. You can describe how your practices align with it, and you should be specific about what you actually do.
Is the AI RMF a law?
No. It is voluntary guidance, though regulators, customers and other frameworks may refer to it.
Sources
- NIST AI Risk Management Framework overview page
- NIST AI 100-1, Artificial Intelligence Risk Management Framework (AI RMF 1.0)
Last reviewed: 18 October 2026.
