CyphrQ
Menu

ISO/IEC 42001 Explained: Who Needs It and How Certification Works

ISO/IEC 42001 is the AI management system standard. What it covers, who should consider it, how audit and certification work, and how it differs from NIST.

By CyphrQ.ai

Written with AI assistance and human editorial review. Every figure linked to its source.

11 min read

ISO/IEC 42001:2023 is an international standard for running AI responsibly as a management discipline. It asks an organisation to set up an AI management system: policies, roles, risk checks and records that cover how it develops, provides or uses AI. An independent auditor can then certify that the system exists and works. It certifies your management system, not any single AI product.

What is ISO/IEC 42001?

Updated 25 October 2026.

ISO/IEC 42001:2023 sets out requirements and guidance for establishing, implementing, maintaining and continually improving an AI management system. ISO, the publisher, lists it as Edition 1, published in December 2023, and calls it "the world's first AI management system standard". The national standards body of Estonia, which sells the adopted text, describes it as intended for organisations that provide or use products or services that use AI systems. Any organisation can use it, whatever its size, type or nature. Estonia's own listing gives 18 December 2023 as the date it came into force there, which is a national date, not ISO's publication date.

A management system is not software. Think of it as the agreed rules, owners and habits that keep an organisation's handling of a topic steady over time. If you have heard of ISO 9001 for quality or ISO/IEC 27001 for information security, this is the same idea applied to AI.

Three words in that definition matter:

  • Requirements: things you must do to claim you meet the standard.
  • Management system: the whole operating routine, not a one-off project.
  • Continually improving: it is a cycle you repeat, not a certificate you earn once and forget.

What does an AI management system actually contain?

Put simply, an auditor will want answers to five questions:

  1. Who is responsible for AI here? Top management sets an AI policy and assigns roles.
  2. What AI do we build, buy or use, and who is affected? You define scope and understand the people and parties involved.
  3. What could go wrong, and what would it do to people? You assess risks to the organisation and the impact of AI systems on individuals and groups.
  4. Which safeguards are in place, and is there proof they work? You pick them, run them, and record that they ran.
  5. How do we check and improve? Internal audits, management reviews and corrective actions.

Structurally, the standard follows the common management-system layout used across ISO's management standards. The requirements cover context, leadership, planning, support, operation, performance evaluation and improvement. If your organisation already runs one ISO management system, the skeleton will look familiar. Clause numbering and exact wording are in the standard itself, which is paid content, so I do not reproduce them here.

On top of that skeleton, the standard adds an annex of AI-specific controls with companion implementation guidance. The annex is paid content, so its control objectives are not listed here. Get the exact list from the standard itself before you plan work around it.

Who needs ISO/IEC 42001, and who can skip it?

Nothing in the standard itself makes certification a legal requirement. Whether it is worth doing depends on who asks you for proof.

Your situation Why 42001 may be worth it
You build or sell AI products or services Some customers ask how you govern AI. A certificate is a recognised answer.
You use AI heavily in regulated or high-stakes work It gives a structure for risk, accountability and records.
You already hold ISO/IEC 27001 You can reuse much of the management-system machinery.
You use a few AI tools lightly A formal certification is probably more than you need. A policy and a tool inventory go a long way.

Does ISO 42001 make you compliant with the EU AI Act?

Not by itself. The Act is a law with dates and duties, and 42001 is a voluntary management standard. The EU AI Act timeline post covers what applies when. Ask your legal adviser what a certificate does and does not show for your case.

How does ISO/IEC 42001 certification work?

The route to a certificate is much the same as for any other ISO management system. Treat the steps below as the general shape; your certification body will give you its own plan and timings.

  1. Scope it. Decide which AI systems, teams and locations the management system covers. Be specific, because the certificate will describe this scope.
  2. Build the system. Write the policy, assign roles, run the risk and impact assessments, choose controls and document them.
  3. Run it and gather evidence. Operate the controls, hold the internal audit and management review, and keep records.
  4. Hire an accredited certification body. A third-party auditor assesses your system against the standard. Choose one that is accredited for this standard by a national accreditation body.
  5. Pass the audit stages and receive the certificate. If the auditor finds gaps, you fix them and show the fix.
  6. Stay certified. Certification does not end at the first certificate. Expect ongoing surveillance audits and a periodic recertification, as with other ISO management systems. Confirm the cycle with your body.

What is ISO/IEC 42006 and why does it matter for 42001 certification?

ISO/IEC 42006:2025 is a companion standard aimed at the auditors. It adds requirements to ISO/IEC 17021-1 for bodies that audit and certify AI management systems under 42001. The Estonian standards body's listing says following it is meant to show that certification bodies are competent, consistent and reliable, and that it can serve as a criteria document for accreditation. That listing gives 7 July 2025 as the date it became valid in Estonia. ISO's own page for it was not checked here, so confirm the publication date and status there.

Why you care: before 42006, "accredited certification" for AI rested on general rules. Now certification bodies have an AI-specific competence yardstick. When you shortlist a body, ask whether it is accredited for 42001 and how it meets 42006.

Who has been certified under ISO/IEC 42001?

Certificates exist, but read each announcement for what it actually covers. Three examples, each from the announcing organisation's or certifier's own words (the Cognizant item is a republished press release, and the "first" is the company's claim):

Organisation Date What the announcement says
Cognizant 16 December 2024 Press release headline: "Cognizant First to Achieve ISO/IEC 42001:2023 Accredited Certification for Artificial Intelligence Management Systems".
Metyis, certified by DNV 1 July 2025 DNV's announcement calls Metyis its first Dutch customer to receive the certification. DNV notes it is accredited by the Dutch Accreditation Council to audit against the standard.
KPMG International, certified by Schellman Announced December 2025 KPMG's release says KPMG International is the first of the Big Four's international entities to earn it, following certifications already held by member firms in four countries: Australia, Spain, India and the USA.

In practice: notice the pattern. Each announcement names an accredited certification body or accreditation, and each claims a "first" within a narrow category. None of these pages define the certified scope in detail. That is the question to ask when a supplier tells you it is "ISO 42001 certified": which AI systems and which business units does the certificate cover?

How does ISO/IEC 42001 differ from the NIST AI RMF?

They solve different problems. NIST's AI Risk Management Framework 1.0 was released on 26 January 2023 and is described as intended for voluntary use. It organises work into four functions: Govern, Map, Measure and Manage. NIST also released a generative AI profile (NIST AI 600-1) on 26 July 2024. On 25 October 2026 NIST's page said: "The AI RMF 1.0 is being revised as part of the White House AI Action Plan." Check the current version.

ISO/IEC 42001 NIST AI RMF 1.0
What it is A management system standard with requirements A voluntary risk management framework
Who publishes it ISO and IEC NIST (US)
Outcome A certificate from an accredited body is possible NIST's page does not describe any certification
Best used for Proving a governed, auditable system to others Structuring how you think about AI risk

They are not rivals. The NIST page describes the RMF as aligning with and supporting other AI risk management efforts. A practical pattern: use the RMF's four functions to organise your thinking, and 42001 as the auditable wrapper around it.

Going deeper: for security and IT teams

Treat it as a sibling of your ISMS, not a replacement. ISO/IEC 27001 manages information security risk. ISO/IEC 42001 manages AI-specific risk and responsible use. Both use the common management-system structure, so many clauses can share the same machinery: document control, internal audit, management review, corrective action and the risk register. Plan one integrated audit programme rather than two. Where they differ is the subject of the risk: 27001 asks what could harm the confidentiality, integrity and availability of information; 42001 also asks what an AI system could do to people and to the organisation's obligations, including bias, transparency and misuse.

Two assessments, two audiences. The standard separates the risk assessment (risk to the organisation and its objectives) from the AI system impact assessment (effect on individuals, groups and society). Do not collapse them. Run the impact assessment per AI system or per material change, record who reviewed it, and feed its outputs into control selection. A security team that owns the risk register will need to bring in legal, privacy and the product owner for the impact side.

Statement of Applicability (SoA). The standard's text was not checked here, so it is not confirmed that 42001 works like 27001 on this point. If it does, you decide which Annex A controls apply, record each as applicable or not, justify exclusions and show implementation status. Ask your certification body whether it expects an SoA. Either way, an exclusion with no reasoning is an easy finding for an auditor. Build any such record from the standard's own text, not a vendor spreadsheet.

What evidence does an ISO 42001 auditor ask for?

Build the evidence list around the five questions earlier:

  • Signed AI policy and role assignments, with a named accountable executive.
  • A current inventory of AI systems you develop, provide or use, including third-party and embedded AI.
  • Risk and impact assessments with dates, owners and approval.
  • Procedures for the AI system life cycle: requirements, testing, release, monitoring, retirement.
  • Data governance records for training and input data where you control them.
  • Supplier due diligence for AI vendors, including what their own certificates cover.
  • Records of your own checks: internal audit reports, minutes from management reviews, and the corrective action log.

Align with existing controls. Map identity and access management, logging, change control and incident response from your ISMS to the AI life cycle rather than writing new ones. Where you use agents or third-party models, the guardrails in AI Agent Security Risks at Work (least privilege, one identity per agent, readable logs, a kill switch) are the kind of operational controls an AI management system should point to.

Choosing the certification body. Verify, on the accreditation body's own register, that the body is accredited for ISO/IEC 42001 and ask how it implements ISO/IEC 42006. Check auditor competence in both AI and your sector. Ask how combined 27001 and 42001 audits are scheduled, how long the certificate cycle runs, and what happens if a major nonconformity is found.

Verify before you rely on a supplier certificate. Ask for the certificate, the scope statement, the certification body and its accreditation, and the SoA summary. If the certificate names one business unit, it proves nothing about the others. Re-check dates and validity, because certificates lapse.

Ask your own team: Which AI systems are in scope and why? Who owns each impact assessment? Could we justify every exclusion in our SoA if challenged?

Where should you start with ISO 42001?

  • List every AI system you build, buy or use. You cannot scope without this.
  • Name one accountable executive for AI and write a short AI policy.
  • Ask your customers and local regulators if they expect evidence, and what sort they want.
  • If you hold ISO/IEC 27001, ask your auditor how 42001 could be integrated.
  • Suppliers claiming certification should show you the scope and name the body that certified them.

Key takeaways

  • ISO/IEC 42001:2023 is a voluntary standard for an AI management system: policies, roles, risk and impact assessment, controls, audit and improvement.
  • It certifies a management system, not a model or a product, and the certificate has a defined scope.
  • ISO/IEC 42006:2025 sets the extra requirements for the bodies that audit and certify against 42001.
  • It is not the same as the NIST AI RMF, which is a voluntary framework; the two work well together.
  • A certificate is useful evidence for customers and auditors, but it is not by itself proof of legal compliance.
  • If you already run ISO/IEC 27001, you have a head start.

Sources