A board asked to adopt an AI governance standard usually hears two names, ISO/IEC 42001 and the NIST AI Risk Management Framework, and assumes it must choose. They do different jobs, and the more useful question is which one you need first.
The short answer
- ISO/IEC 42001 is a certifiable management-system standard. You can be audited against it and hold a certificate, which is worth having when a customer or regulator wants proof.
- The NIST AI RMF is a voluntary framework with no certificate. It is faster to adopt and strong on the practical work of identifying and managing AI risk.
- Neither is legally mandatory in the UK. Adopting one is a way to make the UK principles and, where relevant, the EU AI Act operational.
- Many organisations use the NIST framework to do the risk work and ISO 42001 to prove it, because the two map onto each other cleanly.
- Choose by what you need to demonstrate, to whom, and by when.
What each one actually is
ISO/IEC 42001, published in December 2023, is the first international management-system standard for artificial intelligence. It follows the familiar Plan-Do-Check-Act shape of other ISO management standards, sets requirements for an AI management system, and comes with a set of controls in its annex. Because it is a management-system standard, an accredited body can audit an organisation against it and issue a certificate, typically valid for three years with surveillance in between. That certificate is the point: it is external, recognised, and hard to wave away.
The NIST AI Risk Management Framework, released in January 2023, is a voluntary framework rather than a certifiable standard. It is organised around four functions, Govern, Map, Measure and Manage, that walk an organisation through establishing accountability, understanding context and risk, assessing and tracking that risk, and acting on it. It carries no certificate. An organisation self-attests, and can engage a third party for assurance if it wants outside confirmation, but there is no NIST stamp to hold up.
Where they differ, and where they agree
| ISO/IEC 42001 | NIST AI RMF | |
|---|---|---|
| Type | Certifiable management-system standard | Voluntary risk framework |
| Proof it produces | An accredited certificate after external audit | Self-attestation, or optional third-party assurance |
| Shape | Plan-Do-Check-Act, with annex controls | Govern, Map, Measure, Manage |
| Speed to adopt | Slower; audit preparation and evidence | Faster; start managing risk immediately |
| Best when | You need to demonstrate governance to a third party | You need to get the risk work done well |
| Cost | Certification and surveillance fees, plus the internal work | Internal effort, no certification fee |
The agreement matters as much as the difference. Both put governance and accountability first, both require you to understand and document AI risk, and both are built to be evidenced. Published crosswalks map one to the other, so the risk work done under the NIST functions produces most of the evidence an ISO 42001 audit will ask for. Adopting one does not waste the other.
Neither is the law, and that matters
Neither framework is legally required, in the UK or elsewhere. The UK regulates AI through existing sector regulators applying existing law, as set out in the UK's sector-regulator approach, and the EU does it through the EU AI Act, whose reach into UK companies is covered in who the Act actually binds. A framework is the bridge between those obligations and daily practice. It gives you a structured control set that satisfies the UK principles, supports an EU AI Act conformity case where you are in scope, and travels across regimes so the same system does not need three separate governance stories.
How to choose
The decision comes down to what you must demonstrate, and to whom.
Lead with ISO 42001 when proof is the point
If customers are asking for evidence of AI governance in procurement, if a regulator in your sector expects demonstrable assurance, or if certification would be a differentiator in your market, ISO 42001 is the one that produces the artefact those audiences accept. The certificate answers a question the NIST framework can only answer with your own word.
Lead with NIST when the risk work comes first
If the immediate need is to get a grip on AI risk quickly, to stand up governance without waiting on an audit cycle, or to build the muscle before committing to certification, the NIST framework lets you start on Monday. It is also a sensible first step even when ISO 42001 is the eventual goal, because it front-loads the substance the audit will test.
Use both when you can
For many organisations the honest answer is both, in sequence. Do the risk work with NIST, then certify the management system with ISO 42001 once it is real. Because they map onto each other, the second is far cheaper once the first is done, and the combination gives you competent risk management and the certificate that proves it.
The constraint
The binding limit is whether a named executive owns the AI management system the framework describes, far more than which framework is chosen. Adopting either standard without that owner produces a document set that ages on a shared drive, because both frameworks assume an accountable owner runs the cycle, and no certificate or self-attestation substitutes for the person who is answerable when a system misbehaves.
The question to settle first
Before debating ISO against NIST, answer who will own the AI management system once it exists, and what you most need to prove in the next year. A board that can name the owner and the audience has already made the framework choice easy, because the standard follows from the demonstration you need. A board still arguing about the frameworks in the abstract has skipped the question that actually decides the outcome.
Sources
- ISO/IEC 42001:2023, Information technology, Artificial intelligence, Management system, published December 2023.
- National Institute of Standards and Technology, AI Risk Management Framework (AI RMF 1.0), January 2023.
- NIST, crosswalk material mapping the AI RMF to ISO/IEC 42001, on the overlap between the two.