Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
The NIST AI Risk Management Framework is a voluntary structure for managing AI risk through four functions—Govern, Map, Measure, and Manage—usable by any organisation, anywhere.

What Is the NIST AI RMF?
The AI Risk Management Framework, published by the US National Institute of Standards and Technology in January 2023 as AI RMF 1.0, is a voluntary framework for identifying and managing the risks an AI system creates. It was written to be sector-agnostic, technology-neutral, and usable by an organisation of any size, which is why it has been adopted well outside the United States—including by UK organisations looking for a defensible structure in the absence of domestic AI legislation.
| Function | Question it answers | Typical outputs |
|---|---|---|
| Govern | Who is accountable, and under what policy | Roles, policy, risk tolerance, supplier requirements, culture |
| Map | What is this system, in what context, with what impacts | Inventory, use case documentation, stakeholder and impact analysis |
| Measure | How do we know how it behaves | Metrics, evaluation, adversarial testing, monitoring design |
| Manage | What do we do about what we found | Treatment decisions, controls, incident response, decommissioning |
Govern runs through the other three rather than preceding them, which is the framework’s most useful structural idea: without ownership and policy, mapping and measurement produce documents that nobody acts on.
The framework defines what it is aiming at: systems that are valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. These are explicitly in tension. Tighter privacy protection through differential privacy reduces accuracy; more explainability can constrain model choice; fairness adjustments can lower aggregate performance. The framework does not pretend otherwise. It asks organisations to make the trade-off consciously and record the reasoning, which is also what a regulator or a client questionnaire ultimately wants to see.
In July 2024 NIST published AI 600-1, a profile applying the framework to generative systems. It identifies risks that are specific to or amplified by this class of model, including confabulation, information security risks such as prompt injection, data privacy, harmful bias, intellectual property, and the environmental cost of training. For each it suggests actions mapped back to the four functions, which makes it the more practically useful document for a team deploying language models. Alongside it, NIST AI 100-2 provides the adversarial attack taxonomy and NIST SP 800-226 covers evaluating differential privacy guarantees.
There is no legal obligation to adopt the framework in the UK, and no certification against it. Its value is as scaffolding that maps onto things that do bind you. The Govern function aligns with UK GDPR accountability and with existing information security governance. Map corresponds closely to the scoping stage of a Data Protection Impact Assessment and to building an AI inventory. Measure covers the adversarial testing expected by NCSC and CISA guidance on secure AI development, and described in our note on AI red teaming. Manage connects to incident response and to change control. Organisations that need something certifiable usually pair the framework with ISO/IEC 42001, using the framework for analysis and the standard for the management system. For those in scope of the EU AI Act, the Map and Measure outputs supply much of the technical documentation the Act requires, though the framework is not a conformity route in itself.
Start narrow. Pick the two or three AI systems that carry real consequence, run them through Map properly—context, affected people, plausible failure modes—then define what Measure means for each: the metrics you will track, the adversarial tests you will run, and the thresholds that trigger action. Assign owners under Govern before expanding the inventory, because a framework applied across forty systems with no named owner produces a spreadsheet rather than control. At Privia Security, our Cyber Risk Analysis Services help organisations turn an AI risk framework into assessments and controls that fit how they already work.
No. It is voluntary everywhere, including in the United States, and there is no certification scheme against it. Organisations adopt it because it provides a defensible structure and a shared vocabulary, and because its outputs feed directly into obligations that are mandatory, such as UK GDPR accountability or EU AI Act technical documentation.
The framework is an analytical structure for thinking about AI risk; ISO/IEC 42001 is a certifiable management system standard with requirements for policy, roles, internal audit, and continual improvement. They complement each other, and many organisations run the framework’s Map and Measure activities inside a 42001 management system.
Yes, if applied proportionately. Take the two or three systems where a failure would actually matter, document context and impacts under Map, define a small set of metrics and tests under Measure, and name an owner under Govern. That produces more real control than a comprehensive inventory nobody maintains.
Not by itself. The Act requires specific conformity assessment, technical documentation, logging, and CE marking for high-risk systems, and the framework is not a recognised conformity route. It does generate much of the underlying evidence, so organisations preparing for the Act often use it as the analysis layer beneath a harmonised standard.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.