ISO/IEC 42001

ISO/IEC 42001:2023 is the international standard for AI management systems (AIMS), and the only AI governance standard an organization can be independently audited and certified against. Published on 18 December 2023, it sets requirements for establishing, operating and improving a governed approach to AI — covering policy, roles, risk assessment, impact assessment and third-party relationships. It applies to any organization that develops, provides or uses AI systems.

Certification is what distinguishes it. The NIST AI RMF gives you a risk method with no certificate at the end. ISO/IEC 42001 gives you something an auditor signs.

What the standard actually requires

ISO/IEC 42001:2023 was published by ISO/IEC JTC 1/SC 42. It follows the harmonized management system structure, so clauses 4 to 10 cover context, leadership, planning, support, operation, performance evaluation and improvement in the same order as other ISO management system standards.

The obligations are concrete. Determine your organizational context and your role with respect to AI, since the standard treats developers, providers and users differently, then define and document the scope of the management system. Issue an AI policy from top management and assign responsibilities. Run an AI risk assessment (clause 6.1.2) and an AI risk treatment process (clause 6.1.3). Set measurable objectives, ensure competence, control documented information, run an internal audit program and hold management reviews with defined inputs. Clause 6 plans the risk work; clause 8 requires you to actually perform it.

Annex A is normative and contains reference control objectives and controls, reported as 38 controls across nine objectives covering AI policy, internal organization, resources, impact assessment, life cycle, data, information for interested parties, responsible use, and third-party relationships. A structural detail that distinguishes 42001 from ISO/IEC 27001: Annex B, the implementation guidance, is also normative and sits inside the standard itself. ISO 27001 puts its guidance in a separate document, ISO/IEC 27002.

The impact assessment has no ISO 27001 equivalent

Clause 6.1.4 requires an AI system impact assessment, executed under clause 8.4. This is the requirement with no analogue anywhere in information security management. It assesses consequences for individuals, groups and society, not just risk to the organization. That inversion is the point of the standard: information security asks what could happen to you, and 42001 asks what your AI system could do to other people. ISO/IEC 42005:2025, published in May 2025, is the companion guidance for carrying it out.

How certification actually works

ISO does not certify anyone. Certification is voluntary and carried out by independent certification bodies, which may themselves be accredited by a national accreditation body such as ANAB in the United States or UKAS in the United Kingdom. ISO/IEC 42006:2025, published in July 2025, sets the requirements those certification bodies must meet, adding AI-specific competence requirements on top of ISO/IEC 17021-1.

Two details from 42006 are worth knowing before commissioning an audit. Its Annex A is normative and governs audit time, so duration is formula-driven rather than negotiable. And clause 9 follows the familiar cycle: a two-stage initial certification audit, annual surveillance, and re-certification. A documented worked example exists: AWS certified in November 2024 and completed its first surveillance audit in November 2025 with no findings.

Who has certified

There is no public count of certified organizations. ISO compiles its survey from IAF CertSearch, which is login-gated, and no open register exists. Treat any global total quoted without an IAF CertSearch citation as unsupported, and treat "among the first 30 companies" style claims as marketing, since there is no register against which to check them.

Verifiable certifications include AWS (Amazon Bedrock, Q Business, Textract and Transcribe, November 2024), Microsoft (across GitHub Copilot, Microsoft 365 Copilot, Security Copilot and others, with certificates published on its Service Trust Portal), IBM for its Granite model family, and KPMG International in December 2025.

Relationship to ISO/IEC 27001

The two are designed to share a management system. Both use the harmonized structure, so clauses 4 to 10 line up one for one, and clause 9.1.6 of ISO/IEC 42006 is titled "Multiple management systems", explicitly contemplating combined audits. ISO sells the two as a package.

The overlap has limits. The control sets are entirely different — 42001 Annex A holds AI controls, while ISO 27001 Annex A holds 93 information security controls — and the AI system impact assessment has no counterpart. An existing ISO 27001 program gives you the management system scaffolding and roughly none of the AI-specific content.

What it does not do for the EU AI Act

This is the most consequential misunderstanding about the standard. ISO/IEC 42001 was adopted unchanged as a European Standard, EN ISO/IEC 42001:2026, approved by CEN on 13 March 2026. The 2026 in that designation is a re-badge, not a new edition, and the text is identical to the 2023 standard. There have been no amendments or revisions to ISO/IEC 42001 itself.

Adoption as a European Standard is a text-availability step, not a legal one. Under Article 40 of the EU AI Act, presumption of conformity attaches only to harmonised standards whose references are published in the Official Journal. Tellingly, CEN-CENELEC did not adopt 42001 for the Act’s quality management requirement. It wrote a bespoke standard instead: EN 18286:2026, approved 12 July 2026, described as the first harmonised European standard for AI Act regulatory purposes and tied to the Article 17 obligations on high-risk providers.

The practical position: an ISO/IEC 42001 certificate is strong evidence that you run a governed AI management system, and it is useful to customers, regulators and procurement. It is not a route to presumption of conformity under the EU AI Act, and it does not satisfy Article 17. Anyone selling it as AI Act compliance is overselling it.

The wider standards family

Published and relevant: ISO/IEC 23894:2023 on AI risk management guidance, ISO/IEC 42005:2025 on impact assessment, ISO/IEC 42006:2025 on certification bodies, and ISO/IEC 22989:2022 on concepts and terminology, which is 42001’s only normative reference.

Still in development, and frequently misdescribed as published: ISO/IEC 42003 on implementing 42001, ISO/IEC 42007 on conformity assessment schemes, ISO/IEC 42105 on human oversight, and ISO/IEC 24970 on AI system logging. There is no such standard as ISO/IEC 42002, despite it appearing in vendor content. For how 42001 sits alongside the other frameworks, see the AI security framework comparison and AI risk, audit and compliance.

Related frameworks

ISO/IEC 42001 certifies that a management system exists. For the risk method that populates it with evidence, see NIST AI RMF. For the engineering practice the standard assumes underneath, see OWASP AI. For how all five frameworks relate, see the AI security framework comparison. Putting the management system in place is covered in AI Governance, and AISGRC publishes the artifact set it requires.