ISO 42001 vs NIST AI RMF: Differences, Overlap, and a Practical Crosswalk

ISO/IEC 42001 is a certifiable management system standard for AI. The NIST AI Risk Management Framework is a free, voluntary risk framework that cannot be certified. ISO 42001 tells you what management system to build and gets audited. The AI RMF goes deeper into how to identify, measure, and treat AI risk. They overlap heavily, and a common approach is to use the AI RMF for risk practice and ISO 42001 as the certifiable system around it.

If you are new to either, start with our ISO 42001 guide or our NIST AI RMF guide.

Side-by-side comparison

	ISO/IEC 42001:2023	NIST AI RMF 1.0 (AI 100-1)
Publisher	ISO and IEC (JTC 1/SC 42)	US National Institute of Standards and Technology
Released	December 2023	January 26, 2023
Type	Management system standard with requirements ("shall")	Voluntary framework of outcomes
Certifiable	Yes, by an accredited certification body	No, there is no certification
Cost to obtain	Paid standard from ISO	Free from NIST
Structure	Clauses 4 to 10 plus 38 Annex A controls in 9 areas	4 functions, 19 categories, 72 subcategories
Core risk content	AI risk assessment, risk treatment, AI system impact assessment	MAP, MEASURE, MANAGE, plus 7 trustworthiness characteristics
Management system elements	Scope, policy, objectives, document control, internal audit, management review, corrective action	Governance outcomes under GOVERN, without an audit cycle
GenAI-specific guidance	Applies to all AI; no separate GenAI annex	Generative AI Profile, NIST AI 600-1 (July 26, 2024)
Implementation help	Annex B guidance; ISO/IEC 23894 and 42005	AI RMF Playbook
Integrates with	ISO 27001, ISO 9001 (same harmonized structure)	NIST Cybersecurity Framework, Privacy Framework
What customers get	A certificate valid for 3 years, with annual surveillance audits	Your own description of alignment

The main differences

Certification

This is the difference that matters most in a sales cycle. ISO 42001 certification is issued by an accredited certification body after a Stage 1 and Stage 2 audit. The NIST AI RMF has no certification scheme, and NIST does not audit companies. If a procurement questionnaire asks "Are you certified?", only ISO 42001 can answer yes. See our AI governance certification guide for the full picture.

Requirements vs outcomes

ISO 42001 states requirements an auditor tests. You must have an AI policy, a risk assessment process, a Statement of Applicability, an internal audit, and a management review. The AI RMF describes outcomes, such as "AI risks are prioritized and responded to," and leaves the method to you.

The management system wrapper

ISO 42001 builds a system that runs itself: document control, competence records, internal audit, management review, and corrective action. The AI RMF's GOVERN function covers policy, accountability, and culture, but it does not require an audit and review cycle. This is the biggest gap when you move from the AI RMF to ISO 42001.

Depth of risk guidance

The AI RMF is more detailed on risk. Its seven trustworthiness characteristics, MEASURE subcategories, and the GenAI Profile's 12 risks give teams concrete things to evaluate. ISO 42001 requires risk and impact assessment and points to related standards, but leaves more of the method open.

Practical crosswalk: AI RMF functions to ISO 42001

NIST's AI Resource Center lists a NIST AI RMF to ISO/IEC 42001 crosswalk, contributed by Microsoft, on its crosswalks page. Two caveats. First, NIST says posting a crosswalk does not imply endorsement, and that neither resource necessarily covers the other completely. Second, that crosswalk was built against the draft (FDIS) text of ISO 42001 and cites Annex B guidance numbers, which follow the same numbering as the Annex A controls.

The table below is our simplified, category-level reading of how the two line up. It is approximate. Real mapping happens at the subcategory and control level, and one AI RMF outcome often touches several ISO clauses.

AI RMF category	What it covers	Closest ISO 42001 clauses	Closest Annex A areas
GOVERN 1	Policies, processes, risk tolerance, AI inventory, decommissioning	4.1, 4.4, 5.2, 6.1, 6.2	A.2 Policies, A.4 Resources, A.6 Life cycle
GOVERN 2	Accountability, roles, training, executive responsibility	5.1, 5.3, 7.1 to 7.4, 9.3	A.3 Internal organization
GOVERN 3	Workforce diversity and accessibility in risk work	7.2	A.3, A.4.6 Human resources, A.5
GOVERN 4	Risk-aware culture, documenting and communicating impacts	6.1.4, 7.4	A.5 Impacts, A.6, A.8 Information for interested parties
GOVERN 5	Engagement with external stakeholders and feedback	4.2 (interested parties)	A.5, A.8.3 External reporting, A.10.4 Customers
GOVERN 6	Third-party software, data, and supply chain	8.1 (outsourced processes)	A.10 Third-party and customer relationships
MAP 1	Context, intended purpose, risk tolerance, stakeholders	4.1, 4.2, 6.1.1, 6.2	A.5, A.6.2.2 Requirements, A.9.4 Intended use
MAP 2	Categorizing the system, its tasks, methods, and limits	8.2	A.4 Resources, A.6, A.7 Data
MAP 3	Capabilities, benefits, costs, and human oversight	4.3, 8.2 to 8.4	A.5, A.6, A.8.2 User information
MAP 4	Risks and benefits of all components, including third parties	4.1, 6.1.2	A.10.3 Suppliers, A.9
MAP 5	Likelihood and magnitude of impacts on people and society	6.1.2, 6.1.4, 8.4	A.5 Assessing impacts
MEASURE 1	Choosing methods and metrics	6.1.2, 9.1	A.5, A.6.2.4 Verification and validation
MEASURE 2	Evaluating trustworthiness characteristics	8.2, 9.1	A.6.2.4 to A.6.2.8, A.7 Data
MEASURE 3	Tracking risks over time, including emergent risks	8.2, 10.1	A.6.2.6 Operation and monitoring, A.8.3, A.8.4
MEASURE 4	Feedback on whether measurement works	9.1, 9.2, 9.3	A.6.2.6, A.8
MANAGE 1	Prioritizing and responding to risks	6.1.3, 8.3	A.5, A.6, A.9
MANAGE 2	Maximizing benefits and minimizing harm; deactivation	6.1.3, 7.1, 8.3, 10.2	A.3.3 Reporting of concerns, A.6.2.6, A.7
MANAGE 3	Managing third-party risks	8.1	A.10.2, A.10.3, A.4.4 Tooling
MANAGE 4	Post-deployment monitoring, incidents, communication	9.2, 9.3, 10.2	A.6.2.6, A.8.4 Communication of incidents, A.10.4

Clause references follow the published ISO/IEC 42001:2023 structure. Where our reading differs from the Microsoft crosswalk, it is because we grouped at the category level. Use the official crosswalk for subcategory detail.

What the crosswalk does not cover

Some ISO 42001 requirements have no real AI RMF counterpart. If you are coming from the AI RMF, these are where the work is:

• AIMS scope (4.3) as a formal, auditable boundary
• Statement of Applicability (6.1.3) with a justification for every included and excluded control (see our Annex A controls guide)
• Documented information control (7.5)
• Internal audit program (9.2) with objective auditors
• Management review (9.3) with defined inputs and recorded decisions
• Nonconformity and corrective action (10.2) with root cause analysis

Going the other way, the AI RMF gives more specific direction than ISO 42001 on trustworthiness testing, workforce diversity in risk work (GOVERN 3), and generative AI risks through the GenAI Profile.

Auditor's note: In audits, the most common gap we look for in companies that start from the NIST AI RMF is strong risk content with no operating cycle around it. The risk register is detailed, but nobody has audited it, management has never formally reviewed it, and there is no record of what changed after the last incident. Your AI RMF work is not wasted. It needs to be put inside a system that produces those records.

Using both: a practical approach

For most B2B SaaS companies being asked about AI governance, this sequence works:

1. Build the AIMS to ISO 42001. Define scope, policy, roles, and the audit and review cycle. This is what gets certified.
2. Use the AI RMF inside the risk process. Shape your AI risk assessment and impact assessment around MAP, MEASURE, and MANAGE and the seven trustworthiness characteristics.
3. Apply the GenAI Profile to any generative AI system in scope, and use its 12 risks as prompts in your risk register.
4. Keep one evidence set. Tag each piece of evidence with both the ISO clause or control and the AI RMF subcategory, so one artifact answers both kinds of questions.
5. Describe alignment accurately. Say "ISO/IEC 42001 certified" once you hold the certificate, and "aligned with the NIST AI RMF" for the framework. Do not claim a NIST certification.

If you already run ISO 27001, the ISO 42001 management system pieces will be familiar, because the two standards share the same harmonized structure. Our ISO 42001 vs ISO 27001 and SOC 2 guide covers that integration.

Where to start

If you have AI RMF work in place and need ISO 42001 certification, a gap assessment shows how much of it carries over. TrustAIMS's ISO 42001 gap assessment (from $12,500, about two weeks) maps your current practice, including existing AI RMF artifacts, to clauses 4 to 10 and all 38 Annex A controls. You get a prioritized remediation plan and a draft Statement of Applicability. Our founders are former Big Four auditors. We are not a certification body and cannot certify you, which keeps our advice separate from your audit.

Want a quick read first? Take the free 10-question readiness assessment.