ISO 42001 vs EU AI Act vs NIST AI RMF: What You Actually Need
ISO Cloud Consulting editorial team
ISO 42001 is a voluntary, certifiable management system standard for AI. The EU AI Act is binding law with obligations based on each system's risk. The NIST AI RMF is a free, voluntary US framework for managing AI risk. Most companies must check the law first, then choose ISO 42001 or NIST as their operating model.
Key facts
- ISO/IEC 42001:2023: published December 2023; the first certifiable AI management system standard; 38 Annex A controls.
- EU AI Act: in force since 1 August 2024. Prohibited practices applied from 2 February 2025, general-purpose AI model obligations from 2 August 2025, and Article 50 transparency obligations from 2 August 2026.
- High-risk dates after the 2026 Digital Omnibus: 2 December 2027 for Annex III systems and 2 August 2028 for AI built into regulated products.
- NIST AI RMF 1.0: released January 2023; four functions (Govern, Map, Measure, Manage); voluntary, with a Generative AI Profile (NIST AI 600-1) added in July 2024.
- Link between them: as of 2026, ISO 42001 is not a harmonized standard under the EU AI Act.
How do ISO 42001, the EU AI Act and NIST AI RMF compare?
They answer different questions. The AI Act says what you must do by law in the EU. ISO 42001 says how to run a management system you can certify. The NIST AI RMF offers a flexible way to think through AI risk.
| Question | ISO 42001 | EU AI Act | NIST AI RMF |
|---|---|---|---|
| What is it? | International management system standard | EU regulation (law) | US government framework |
| Mandatory? | No, unless a customer contract requires it | Yes, where it applies | No |
| Who is it for? | Any organization developing, providing or using AI | Providers, deployers, importers and distributors of AI in the EU market | Any organization; widely referenced in the US |
| How is it structured? | Clauses 4 to 10 plus 38 Annex A controls | Obligations by risk tier and by role | Govern, Map, Measure, Manage |
| What proof do you get? | Certificate from an accredited body | Conformity assessment and documentation for high-risk systems | None formal; self-declared alignment |
| Cost to access | Standard is purchased; certification is paid | Free to read; compliance cost varies by role and risk | Free |
What does the EU AI Act require, in brief?
The EU AI Act sorts AI systems by risk and sets obligations accordingly: some practices are banned, high-risk systems carry heavy requirements, certain systems must be transparent, and most other AI is lightly regulated.
- Prohibited practices: a short list of uses considered unacceptable, banned since 2 February 2025.
- High-risk systems: AI in areas listed in Annex III, such as employment and access to essential services, and AI built into regulated products under Annex I. Providers face requirements for risk management, data governance, technical documentation, logging, human oversight and a quality management system.
- Transparency (Article 50): from 2 August 2026, for example telling people they are interacting with AI and marking certain AI-generated content.
- General-purpose AI models: obligations on model providers since 2 August 2025.
Your obligations also depend on your role. A provider that builds and markets a system carries more than a deployer that uses one. The 2026 Digital Omnibus changed several dates and softened the AI literacy duty in Article 4, so check the current text before relying on any summary, including this one.
What is the NIST AI RMF, in brief?
The NIST AI Risk Management Framework is a voluntary guide for identifying, measuring and managing AI risk, organized into four functions. It is flexible, free and practical, but it does not lead to a certificate.
- Govern: culture, policies, roles and accountability for AI risk.
- Map: understand the context, purpose and potential impacts of each AI system.
- Measure: test, evaluate and track the risks you mapped.
- Manage: prioritize, treat and monitor risks, and respond to incidents.
Engineers tend to like it because it reads like a sensible risk workflow rather than an audit checklist. NIST publishes the framework and companion resources on its AI RMF page.
Where do the three overlap?
They overlap heavily on the fundamentals: knowing your AI systems, assessing risk and impact, documenting and testing systems, informing users and monitoring in use. Work done once can serve all three.
| Activity | ISO 42001 | EU AI Act (high-risk and transparency) | NIST AI RMF |
|---|---|---|---|
| Governance and roles | Clause 5, A.2, A.3 | Provider and deployer duties, quality management | Govern |
| Inventory and context | Clause 4, A.4 | Classifying systems by risk tier and role | Map |
| Risk assessment | 6.1.2, 8.2 | Risk management system | Map, Measure |
| Impact on people | 6.1.4, A.5 | Fundamental rights impact assessment for certain deployers | Map |
| Testing, documentation, logging | A.6, A.7 | Technical documentation, data governance, record-keeping | Measure, Manage |
| Information for users | A.8 | Transparency and instructions for use | Govern, Manage |
| Monitoring and improvement | Clauses 9, 10, A.6.2.6 | Post-market monitoring, incident reporting | Manage |
Which one does your company actually need?
You need the EU AI Act wherever it applies to you, because it is law. Beyond that, choose ISO 42001 if customers want independent proof, and use the NIST AI RMF if you want a free, flexible method without certification.
- Check the law first. Do you place AI on the EU market, use AI in the EU, or produce output used there? If so, identify your role and each system's risk tier.
- Listen to your customers. If enterprise buyers ask for certification or send detailed AI questionnaires, ISO 42001 gives you a recognized answer.
- Pick an operating model. ISO 42001 if you want a certificate or already run ISO 27001; the NIST AI RMF if you want lighter, uncertified structure, common in US public-sector conversations.
- Use the others as inputs. Record AI Act obligations as legal requirements inside ISO 42001 (clause 4.2), and borrow NIST's Map and Measure questions to strengthen your risk and impact assessments.
- If AI use is limited to staff tools, start with an acceptable use policy and an inventory, and revisit when AI enters your product.
What does this look like for real companies?
Three fictional examples show how the answer changes with your role and your AI.
| Company (fictional) | Situation | What they need |
|---|---|---|
| Fieldstone Analytics | US software firm selling an AI demand-forecasting tool to EU retailers, with an AI chat assistant in the product | AI Act provider duties, including Article 50 transparency for the assistant; ISO 42001 because enterprise buyers ask for proof |
| Maple Street HR Software | Sells an AI tool that ranks job applicants to employers in the EU | Likely high-risk under the Annex III employment area, so heavy obligations from 2 December 2027; ISO 42001 as the backbone for meeting them |
| Ridgeline Machining | 40-person US manufacturer whose staff use office copilots and chat assistants | An acceptable use policy and a tools inventory now; NIST AI RMF questions if AI use grows |
Can one program cover all three?
Yes. The most efficient approach is to run ISO 42001 as the management system, treat EU AI Act obligations as requirements within it, and use the NIST AI RMF as a practical toolkit for finding and measuring risk.
In practice this means one AI system inventory with extra fields for AI Act role and risk tier, one risk method that borrows NIST's categories, one impact assessment template that can be extended into a fundamental rights assessment where needed, and one audit and review cycle. The ISO 42001 AIMS Implementation Toolkit ($249) gives you the core documents to build that single program. For the impact piece, see our AI impact assessment guide.
What mistakes do companies make when comparing them?
The most expensive mistake is assuming an ISO 42001 certificate equals AI Act compliance. It helps a great deal, but legal obligations must be checked system by system.
- Assuming the AI Act only affects EU-based companies.
- Claiming to be "NIST certified" for AI; no such certificate exists.
- Waiting for the later high-risk dates and missing the Article 50 transparency obligations that apply from 2 August 2026.
- Running three separate projects with three inventories that never agree.
New to the standard? Read what ISO 42001 is, or see the full ISO 42001 guide and templates. This article is general information, not legal advice.
Last reviewed: 29 September 2026