ISO/IEC 42001

The standard text is not freely available. ISO/IEC 42001:2023 is sold, not published openly: 51 pages, CHF 225 for a single-user electronic copy from the IEC Webstore, with equivalent national adoptions priced separately by BSI, ANSI, SAI, SIS and others. This page therefore describes the standard’s scope, structure and role from ISO’s and IEC’s public catalogue entries and from public accreditation-body material. It does not reproduce, paraphrase or summarise clause text, because that text is not visible without a licence. Anything on this page that reads like a clause requirement should be checked against the purchased standard before you rely on it.

What an AI management system is

ISO/IEC 42001 does not certify a model, a product or an output. It certifies a management system: in ISO’s public phrasing, a set of interrelated or interacting elements of an organisation intended to establish policies and objectives, and processes to achieve those objectives, in relation to the responsible development, provision or use of AI systems.

The published scope statement, carried identically across the IEC, ANSI, SIS and Standards New Zealand catalogue entries, is:

This document specifies the requirements and provides guidance for establishing, implementing, maintaining and continually improving an AI (artificial intelligence) management system within the context of an organization. This document is intended for use by an organization providing or using products or services that utilize AI systems. […] This document is applicable to any organization, regardless of size, type and nature, that provides or uses products or services that utilize AI systems.

Three consequences follow from that wording and are worth being explicit about.

It covers both providers and users. Unlike a product standard, it addresses the organisation that uses AI systems as much as the one that builds them. An enterprise deploying purchased models is squarely in scope.

It contains requirements and guidance in one document. “Specifies the requirements and provides guidance” means the certifiable requirements sit alongside informative material. Only the requirements are auditable, and telling them apart requires the text.

It is organisation-scoped, not system-scoped. Certification says an organisation runs a functioning AI management system across a declared scope. It says nothing about whether any particular model is accurate, fair or safe. This is the same relationship ISO/IEC 27001 has to any individual server, and it is the most frequently misread aspect of the standard.

Catalogue facts

At the time of writing, the IEC Webstore history table lists no superseding edition, so 42001:2023 edition 1.0 is the current text. ISO describes it publicly as the world’s first AI management system standard.

The Annex SL structure

ISO/IEC 42001 is a management system standard, so it is built on the common structure that ISO requires of all such standards. That structure was historically known as Annex SL and is now the harmonised structure of the ISO/IEC Directives Part 1. It is why the standard is familiar on sight to anyone who has implemented ISO 9001 or ISO/IEC 27001: the top-level clause numbering and headings are shared across the family, so only the AI-specific content differs.

The left-hand columns are the harmonised structure common to all ISO management system standards. The right-hand column reflects public descriptions of what 42001 adds; it is not drawn from the clause text.

The practical value of the shared structure is integration, not novelty. An organisation with a running ISO/IEC 27001 information security management system already has clauses 4, 5, 7, 9 and 10 substantially in place: the risk assessment methodology, the internal audit programme, the management review cadence, the documented-information controls. Extending scope to cover AI is a smaller project than starting one, and certification bodies routinely conduct integrated audits.

What does not carry over is the risk framing. Information security risk is risk to the organisation, expressed over confidentiality, integrity and availability. AI risk under this family of standards includes risk to individuals and society arising from the organisation’s systems. Reusing a 27001 risk register wholesale, with its asset-centric scoring, is the most common implementation mistake. The result contains no entries for the harms the standard exists to address.

ISO/IEC 42001 sits within a wider SC 42 portfolio that includes ISO/IEC 23894 on AI risk management guidance and ISO/IEC 42005 on AI system impact assessment. Those are separately purchasable and separately paywalled; NIST hosts community crosswalks from the AI RMF to both.

How certification works

Certification to ISO/IEC 42001 follows the standard third-party conformity assessment route used across ISO management system standards.

1. Implement. Define the scope, establish the management system, run it long enough to generate records, typically at least one cycle of internal audit and one management review.
2. Stage 1 audit. A readiness and documentation review by the certification body: is the scope coherent, is the documented system in place, is the organisation ready for stage 2.
3. Stage 2 audit. Assessment of the system in operation against the standard’s requirements, sampling evidence.
4. Certification decision. Taken by the certification body, separately from the audit team.
5. Surveillance and recertification. Periodic surveillance audits during the certificate cycle, then recertification.

Two roles must be distinguished. The certification body audits and issues certificates. The accreditation body - UKAS, ANAB, SCC and their national equivalents - assesses the certification body. An unaccredited certificate is issued against the same standard but carries no external check on the auditor’s competence. Buyers who look closely will ask which accreditation body’s mark appears on it.

ISO/IEC 42006:2025 is the standard that governs the auditors. It specifies additional requirements to ISO/IEC 17021-1 for bodies performing audit and certification of AI management systems according to ISO/IEC 42001, covering competence criteria for each certification role, audit time calculation, and certification documents. The Standards Council of Canada’s transition bulletin records that the published version added or clarified requirements on liability (5.3.2), audit time calculation (9.1.4.2 and Annex A), competence of personnel involved in certification activities (7.1, 7.2), certification documents (8.2.2) and access to the organisation’s documentation (8.4.2), and that new applications and scope extensions are assessed against the published 2025 version.

The consequence for a buyer of certification services is timing: accredited AIMS certification depends on the certification body itself holding accreditation against ISO/IEC 42006:2025. Ask prospective certification bodies for their accreditation scope and its status against that standard before signing. Check the accreditation body’s public directory, not the certification body’s own claim.

Relationship to the NIST AI RMF

The two are complementary in form and largely overlapping in substance.

Neither replaces the other and neither confers the other’s benefit. The AI RMF gives a richer vocabulary for reasoning about a specific system: trustworthiness characteristics, the map/measure split, the explicit requirement to record what cannot be measured. ISO/IEC 42001 gives an auditable organisational wrapper and a certificate a customer can be shown.

In practice teams run the AI RMF, or something like it, as the per-system process that produces the records an ISO/IEC 42001 audit then samples. NIST’s crosswalk page lists a Microsoft-provided crosswalk from the AI RMF to ISO/IEC 42001, alongside crosswalks to ISO/IEC 23894 and ISO/IEC 42005. NIST attaches an explicit caveat to that entire page: inclusion “does not imply NIST endorsement of the resource the AI RMF is mapped to, nor does it imply that either resource comprehensively covers the contents of the other.” A crosswalk aligns vocabulary. It does not establish equivalence, and no mapping exercise converts AI RMF work into a certificate or a certificate into AI RMF conformance.

Relationship to the EU AI Act

This is where the most damaging claims get made, so it is worth stating plainly.

ISO/IEC 42001 certification is not conformity with the EU AI Act.

The Act establishes its own conformity route for high-risk AI systems: the Chapter III Section 2 requirements, conformity assessment under Article 43, an EU declaration of conformity under Article 47, CE marking under Article 48 and registration under Article 49. Presumption of conformity arises under Article 42 and Article 40 from harmonised standards, meaning standards developed under a Commission standardisation request and cited in the Official Journal of the European Union. ISO/IEC 42001 is an ISO/IEC International Standard, not by itself a European harmonised standard cited in the OJ, and it is a management system standard rather than a product-requirements standard. Its scope statement does not claim conformity with any regulation.

There are also structural mismatches. The Act’s obligations attach per AI system and per role (provider, deployer, importer, distributor); certification attaches per organisation and per declared scope. The Act’s high-risk requirements are specific: data governance under Article 10, human oversight under Article 14, accuracy, robustness and cybersecurity under Article 15. None of them are discharged by evidence that a management system exists.

What certification can honestly do:

• Supply the process substrate. Article 17 requires providers of high-risk systems to operate a quality management system. A functioning AIMS produces much of the same documented process, and overlaps with the risk management system required under Article 9.
• Generate the records. Documented information, internal audit results, management review minutes and corrective action records are the kind of evidence an authority requesting documentation is likely to want to see.
• Answer procurement questions. A certificate is a fast, third-party-verified answer to “do you govern your AI”, which is often what a customer questionnaire is actually asking.

What it cannot do: substitute for conformity assessment, create a presumption of conformity, or support a public statement that a product is EU AI Act compliant. If you need that claim, the route runs through Chapter III of the Regulation.

Whether to pursue it

Certification is worth it when an external party needs to be convinced and will not accept self-attestation: enterprise or public sector procurement, or a contractual requirement. It is also worth it when an organisation needs a forcing function to make AI governance durable rather than a one-off exercise, because surveillance audits impose a cadence that internal intent usually does not.

It is not worth it when the goal is to improve how a specific model behaves. Nothing in a management system standard evaluates a model. For that, run the per-system work. And note that the standard being paywalled means the implementation cost includes buying the text for everyone who needs to read it, which for a distributed engineering organisation is not a rounding error.

Related

• NIST AI Risk Management Framework - free, voluntary, per-system risk framework
• EU AI Act - binding EU regulation with its own conformity route
• AI governance - index