NIST AI RMF vs ISO 42001 vs EU AI Act: how they compare and overlap

A crosswalk of the shared AI governance controls, the EU AI Act dates as amended, and which framework to start with.

12 min read
In short: The NIST AI RMF is voluntary guidance, ISO/IEC 42001 is an international standard for an AI management system, and the EU AI Act is law. The NIST AI RMF and the EU AI Act share controls for inventory, ownership, purpose, third-party providers, monitoring, and logs, so evidence collected once at the control counts toward both. Start with NIST for structure, ISO if customers ask for it, and the AI Act if you have EU exposure.

How do NIST AI RMF, ISO 42001, and the EU AI Act compare?

The NIST AI RMF is voluntary guidance. ISO/IEC 42001 is an international standard for an AI management system. The EU AI Act is law. The NIST AI RMF and the EU AI Act overlap on inventory, accountability, third-party risk, monitoring, and records, so work on those shared controls can count toward both, and toward the management system that ISO/IEC 42001 describes. They differ in legal force, scope, and how much documentation they expect.

This article compares the three side by side, lists the controls they share, and explains which one to start with. For what AI governance covers in general, see what is AI governance. Framework references use each framework's own identifiers with short titles of our own. They paraphrase the frameworks and don't quote them. ISO/IEC 42001 is sold rather than published openly, so this article describes it only from its publisher's catalogue entry and doesn't cite its clauses or controls.

What are the main differences between the three frameworks?

The main differences are legal force, structure, and who the framework applies to. The following table compares them on those points.

AttributeNIST AI RMF 1.0ISO/IEC 42001:2023EU AI Act
TypeRisk management frameworkAI management system standardRegulation, Regulation (EU) 2024/1689
Published byUS National Institute of Standards and Technology (NIST)

International Organization for Standardization (ISO) and International Electrotechnical Commission (IEC), through their joint committee ISO/IEC JTC 1/SC 42, in December 2023

European Parliament and Council of the European Union
Legal forceVoluntaryA standard, not a lawBinding law, with fines for breaches
Who it applies toAny organization that chooses to use it

Any organization, of any size, that provides or uses products or services that use AI systems

Providers, deployers, importers, and distributors of AI systems placed on the EU market or used in the EU, including some outside the EU

StructureFour functions: GOVERN, MAP, MEASURE, MANAGE

Requirements to establish, implement, maintain, and continually improve an AI management system

Prohibited practices (Article 5), high-risk systems (Chapter III), transparency duties (Article 50), and general-purpose AI models (Chapter V)

Documentation weightAs much as you decide is usefulSet by the standard's requirements

Light for systems that aren't high-risk. Heavy for high-risk providers: technical documentation, logs, a quality management system, and conformity assessment

The NIST AI RMF organizes risk work into four functions, each split into categories and subcategories. NIST also publishes a Playbook with suggested actions for each subcategory, available from its AI Resource Center, and a Generative AI Profile (NIST AI 600-1).

ISO/IEC 42001 specifies requirements and guidance for establishing, implementing, maintaining, and continually improving an AI management system within an organization. Its publishers describe it as intended for organizations that provide or use products or services that use AI systems, of any size, type, or nature, and as a way to help them develop, provide, or use AI responsibly.

The EU AI Act sorts AI systems by risk and puts most duties on high-risk systems. It separates the provider, who develops a system and places it on the market under its own name, from the deployer, who uses it. A software company can be both: a deployer of third-party AI and a provider of the AI features it ships. For the deployer side, see EU AI Act for deployers.

When do the EU AI Act's obligations apply?

The EU AI Act applies in stages, and an amending regulation moved the high-risk dates. The following dates come from Article 113 of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744, in the consolidated text on EUR-Lex dated July 27, 2026, checked on November 16, 2026.

FromWhat applies
February 2, 2025

General provisions, including AI literacy (Article 4), and most prohibited practices (Article 5)

August 2, 2025

Rules for notified bodies and general-purpose AI models, governance bodies, and most penalties

August 2, 2026

The rest of the Regulation unless listed otherwise, including the transparency duties in Article 50

December 2, 2027

High-risk requirements for systems in Annex III areas, including deployer duties in Article 26

August 2, 2028

High-risk requirements for AI in products covered by the EU legislation listed in Annex I

Regulation (EU) 2026/1744 changed these dates, and further changes are possible. Confirm the current consolidated text on EUR-Lex before you plan against a date.

Which controls do the NIST AI RMF and the EU AI Act share?

The frameworks share controls for inventory, ownership, purpose, third-party providers, data, human oversight, monitoring, logs, and retirement. The following crosswalk maps each shared control to its NIST AI RMF and EU AI Act references. It leaves out ISO/IEC 42001, whose clause and control text is sold rather than published openly. EU AI Act references apply to high-risk systems unless the row says otherwise.

Shared controlNIST AI RMFEU AI Act
Inventory of AI systemsGOVERN 1.6

No general inventory duty. Article 49 registers some high-risk systems in the EU database

Named owner and rolesGOVERN 2.1Article 26(2), human oversight by competent people
Documented purposeMAP 1.1Article 26(1), use according to instructions
Risk classificationGOVERN 1.3, MAP 2Article 6 and Annex III
Periodic reviewGOVERN 1.5Article 9, for providers, across the life cycle
Third-party AI providersGOVERN 6.1, MANAGE 3.1Article 26, for systems a deployer obtains from a provider
Data the system handlesMEASURE 2.10Article 26(4), input data fits the purpose
Human oversightMAP 3.5Article 14 for providers, Article 26(2) for deployers
Monitoring in productionMEASURE 2.4, MANAGE 4.1Article 26(5) for deployers, Article 72 for providers
Event logs and retentionMANAGE 4.1

Article 12 (logging by design), Article 19 and Article 26(6) (keep logs at least six months)

RetirementGOVERN 1.7, MANAGE 2.4

No general duty. Article 26(5) requires a deployer to suspend use of a system that presents a risk

For the item-level view of each framework, use the NIST AI RMF checklist, the ISO 42001 Annex A checklist, and the EU AI Act checklist. The AI governance checklist combines the shared items into one list.

Where do the three frameworks ask for different things?

The frameworks diverge on what sits outside the shared controls. Each one has work that the others don't ask for.

  • NIST AI RMF goes furthest on measurement. Its MEASURE function asks you to test validity, safety, fairness, explainability, privacy, and environmental impact, and to track risks that are hard to measure. It doesn't tell you how much of this to do; you set that by your risk tolerance.
  • ISO/IEC 42001 asks for a management system, not only controls: requirements to establish, implement, maintain, and continually improve how an organization governs AI. Check the specific requirements against your own copy of the standard.
  • The EU AI Act asks for specific legal duties tied to a system's risk level and your role. Providers of high-risk systems carry most of them, including technical documentation, a quality management system, conformity assessment, and post-market monitoring. Deployers of high-risk systems keep logs, assign oversight, and inform workers and affected people. Every provider and deployer takes measures to support AI literacy under Article 4.

Which framework should you start with?

Start with the NIST AI RMF for structure, add ISO/IEC 42001 if customers ask for it, and work through the EU AI Act if your AI systems are on the EU market or their output is used in the EU. Many teams use more than one, and the shared controls in the crosswalk do the most work across them.

The following table matches common situations to a starting point.

Your situationStart withWhy
No AI governance program yetNIST AI RMF

Free, flexible, and organized by function, so it gives a structure to fill in

Customers ask for ISO/IEC 42001ISO/IEC 42001

It's the international standard for an AI management system, for organizations of any size that provide or use AI

EU customers, users, or staffEU AI Act

It's law; classify your systems and your role before anything else

You ship AI agents and use coding agentsThe shared controls

Inventory, owners, provider decisions, and logs count toward more than one framework

How can one evidence set count toward several frameworks?

One evidence set counts toward several frameworks when you collect evidence once, at the control, and map it to each framework's references afterward. The record that a system has an owner is the same record whether an assessor reads it as GOVERN 2.1 or Article 26(2).

This works because evidence describes what happened, and frameworks describe what to look for. An inventory, an owner per system, a provider decision with its date, a review date, and runtime decision records are facts about your systems. A mapping layer then lists every requirement in each framework and marks which facts support it. When a framework changes, you change the mapping, not the evidence.

Keep the mapping honest by listing requirements that nothing supports. An evidence report that names only the covered rows looks complete and isn't. A report that lists every requirement and marks each one as supported or not covered shows the assessor where your own process has to fill the gap.

What does Arcjet not cover?

Arcjet supplies evidence for some requirements in each framework, and it doesn't make an organization compliant with any of them. The following work stays with your organization:

  • NIST AI RMF: policies, risk tolerance, training, impact assessments with affected people, and measurement of validity, fairness, and explainability.
  • ISO/IEC 42001: the AI management system itself, which your organization establishes, runs, and improves.
  • EU AI Act: legal classification, AI literacy, prohibited-practice assessment, transparency notices under Article 50, fundamental rights impact assessments under Article 27, and reporting to providers and authorities.
  • Any AI system that Arcjet doesn't observe, such as tools on unmanaged devices or AI features inside SaaS products that make no separate connection.

Where Arcjet fits

Arcjet collects one evidence set from what it observes and enforces, and maps it to NIST AI RMF 1.0, ISO/IEC 42001, and the EU AI Act on its evidence page. You can add frameworks of your own in the same shape. The page lists every requirement in each framework and marks each one as supported or not covered, and each supported requirement drills down to the records behind it.

The records come from the product. The AI inventory groups each actor with the providers and MCP servers it uses into an AI system, with the share of enrolled devices that are reporting. Each AI system carries an owner, a purpose, declared data classes and tools, a risk tier, an EU AI Act role, and a review date. Provider decisions are enforced on managed macOS and Windows devices, and use after a prohibition is a violation attributed to a person. Guard decision records, Hold decisions routed to the owner, and opt-in long-term retention cover monitoring and logs.

The evidence page rolls up every standard project in a team. Export it as JSON or CSV, or push it to Vanta or Drata. There's no point-in-time recompute, so the export is the record. For the procedure, see how to export AI governance evidence.

Frequently asked questions

Does ISO 42001 certification satisfy the EU AI Act?

No. ISO/IEC 42001 is a standard for an AI management system. The EU AI Act sets legal duties tied to each system's risk level and your role as provider or deployer, and following a standard doesn't replace them. A working management system can make those duties easier to meet and document.

Is the NIST AI RMF mandatory?

No. NIST states that the AI Risk Management Framework is intended for voluntary use. Customers can still ask about it in security questionnaires.

Can I be certified against the NIST AI RMF?

The AI RMF is voluntary guidance, and NIST's AI RMF page describes no certification program. You assess yourself against its functions and subcategories, often using the NIST AI RMF Playbook.

When do the EU AI Act's high-risk obligations apply?

Under Article 113 as amended by Regulation (EU) 2026/1744, high-risk requirements apply from December 2, 2027 for systems in Annex III areas, and from August 2, 2028 for AI in products covered by Annex I legislation. These dates come from the EUR-Lex consolidated text dated July 27, 2026. Confirm the current text before planning against them.

Does the EU AI Act apply to companies outside the EU?

It can. The Act applies to providers that place AI systems on the EU market and to deployers in the EU, and it also reaches providers and deployers outside the EU when the output of their AI system is used in the EU. Check your position with legal advice.

AI runtime security in your code

Protect your AI agent workflows with Arcjet

Collect one evidence set from what Arcjet observes and enforces, and see it mapped to NIST AI RMF, ISO/IEC 42001, and the EU AI Act.