NIST AI RMF: a practical structure for governing AI systems
The AI Risk Management Framework is the reference most US organisations, regulators and procurement teams now use to ask whether AI is being deployed responsibly. It is voluntary and outcome-based, like the Cybersecurity Framework, and it is the fastest way to put a defensible governance structure around AI use before a customer, auditor or regulator asks. This guide explains the functions, the trustworthiness characteristics and how to operationalise them.
The framework treats AI risk as a socio-technical problem: harms to people, organisations and ecosystems that arise from how a system is designed, trained, deployed and used, not only from software defects. It asks organisations to govern AI deliberately, map the context and risks of each system, measure them with appropriate metrics and manage them through the system’s life.
The Generative AI Profile added in 2024 identifies twelve risks specific to generative models, including confabulation, data privacy, information integrity, intellectual property, harmful bias and value-chain integration, and suggests more than two hundred actions against the AI RMF subcategories. It has become the de facto checklist for enterprise generative AI programmes in the United States.
The four functions
Govern
Policies, roles, accountability, risk tolerance and culture for AI across the organisation; an inventory of AI systems; processes for third-party AI and for decommissioning. Govern is cross-cutting and informs the other three.
Map
Establish the context for each AI system: intended purpose and uses, stakeholders, benefits, costs, the specific risks and their likelihood and impact, and the limits of the data and the model.
Measure
Quantitative and qualitative methods to assess, benchmark and monitor the risks and the trustworthiness characteristics, including testing in conditions similar to deployment, tracking of metrics and independent review.
Manage
Prioritise and respond to risks based on their assessed impact: mitigate, transfer, avoid or accept; plan for incidents and for recovery; monitor post-deployment; communicate with affected parties.
Who it applies to
Any organisation that develops, procures or deploys AI systems, from a bank using models for credit decisions to a company rolling out generative AI assistants to staff. US federal agencies are directed to risk management consistent with it; several state laws on automated decision-making reference it; and enterprise customers, insurers and investors increasingly ask vendors whether they follow it. Organisations pursuing ISO/IEC 42001 certification or EU AI Act compliance use it as the risk-management layer underneath.
From scope to evidence, in one system
AI system inventory as the Govern anchor
Every AI system, internal or third-party, is recorded with its purpose, owner, data, model type, users and risk classification, which is the inventory the Govern function and most AI laws require.
Map and Measure as structured assessments
Each system is assessed against the AI RMF subcategories and the trustworthiness characteristics, with risks scored in the Risk & Quantification module and metrics and test results attached as evidence.
Generative AI Profile actions as controls
The AI 600-1 suggested actions are carried as controls with owners and status, so a generative AI programme has a checklist that maps to the framework rather than a slide deck.
AI policies written to the framework
The AI Policy Builder generates the acceptable-use, development, procurement and oversight policies the Govern function expects, versioned and acknowledged by staff.
Crosswalk to ISO/IEC 42001 and the EU AI Act
AI RMF outcomes are mapped to ISO 42001 controls and EU AI Act obligations, so one assessment supports a certification, a regulatory file and a customer questionnaire.
Frequently asked questions
Is the AI RMF mandatory?
No. It is voluntary, but it is referenced in US federal guidance, in state automated-decision laws and in procurement, and customers ask for it by name. Following it is the simplest way to show due diligence.
AI RMF or ISO/IEC 42001?
They complement each other. The AI RMF is a risk-management framework with no certification; ISO 42001 is a certifiable management-system standard. Most organisations use the AI RMF to structure risk work and ISO 42001 when they need a certificate.
What is in the Generative AI Profile?
Twelve risks specific to generative AI and more than two hundred suggested actions mapped to AI RMF subcategories. It is the practical checklist for organisations deploying large language models, whether built in-house or bought.
How does it relate to the EU AI Act?
The Act is law with obligations by risk tier; the AI RMF is a method. NIST publishes a crosswalk, and the AI RMF’s Map, Measure and Manage work produces much of the technical documentation and risk-management evidence the Act requires for high-risk systems.
Related frameworks
Reviewed November 2026. This guide is general information about the published framework and is not legal advice; always check the current official text from the publisher before relying on it.