ISO 42001: what it is and how it relates to the AI Act

ISO/IEC 42001 is the international standard for an AI management system: the policies, roles, risk and impact assessments, controls and evidence an organisation uses to govern how it develops, provides or uses AI. ISO and IEC published the first edition in December 2023. It is certifiable by an accredited third party, but it is not a harmonised standard under the EU AI Act and does not give presumption of conformity.

Most people searching for ISO 42001 have three questions: what the standard actually asks for, how it relates to the AI Act (Regulation (EU) 2024/1689), and how much of an existing ISO/IEC 27001 programme can be reused. This guide takes them in that order, with references you can cite in an audit or a supplier questionnaire.

What ISO/IEC 42001 is

The full title is Information technology, Artificial intelligence, Management system. It was prepared by the joint technical committee ISO/IEC JTC 1, subcommittee SC 42, which covers artificial intelligence. In Europe it is also published as EN ISO/IEC 42001:2026, the European adoption of ISO/IEC 42001:2023, published by CEN on 18 March 2026 under the joint committee CEN/CLC/JTC 21. National versions carry their own prefixes, such as BS EN ISO/IEC 42001:2026 in the United Kingdom and I.S. EN ISO/IEC 42001:2026 in Ireland (NSAI Standards Shop catalogue).

The scope is broad on purpose. The standard specifies requirements and guidance for establishing, implementing, maintaining and continually improving an AI management system, and its abstract states that it is applicable to "any organization, regardless of size, type and nature, that provides or uses products or services that utilize AI systems".

Two consequences follow, and they surprise many first-time readers:

• It is not only for organisations that build models. A company that buys an AI service and puts it into a business process is in scope too. Your role (developer, provider, user) changes which requirements and controls apply, not whether the standard applies.
• It is not a technical standard for the model. It does not prescribe an architecture or an accuracy target. It asks who decides, how risks and impacts were assessed, and what evidence shows it.

The logic is the same as ISO/IEC 27001: a risk-driven management system, continually improved and verifiable by an independent auditor. What changes is the subject being governed.

Who it is relevant to in the mid-market

• Software companies with AI features whose customers ask how those features are governed.
• Organisations using AI in processes that affect people: recruitment, credit decisions, customer service, clinical or industrial decision support.
• Suppliers to regulated sectors, where AI questions now sit in vendor questionnaires next to information security.

How the standard is structured

ISO/IEC 42001 follows the same layout as the other ISO management system standards. The requirements sit in clauses 4 to 10, followed by four annexes.

• Clause 4, Context of the organisation. Internal and external issues, interested parties, and a documented scope for the AI management system.
• Clause 5, Leadership. Top management commitment, an AI policy, assigned roles and responsibilities.
• Clause 6, Planning. The core of the standard: AI risk assessment (6.1.2), AI risk treatment (6.1.3), AI system impact assessment (6.1.4), plus objectives and planning of changes.
• Clause 7, Support. Resources, competence, awareness, communication, documented information.
• Clause 8, Operation. The same three activities applied in day-to-day work: risk assessment (8.2), risk treatment (8.3) and impact assessment (8.4) recur, they are not a one-off exercise.
• Clause 9, Performance evaluation. Monitoring, internal audit programme, management review.
• Clause 10, Improvement. Continual improvement, nonconformity and corrective action.

The four annexes

• Annex A (normative): reference control objectives and controls.
• Annex B (normative): implementation guidance for those controls. This is the part to read when you need to know what applying a control means in practice.
• Annex C (informative): potential AI-related organisational objectives and risk sources, a useful starting point for the analysis.
• Annex D (informative): use of the AI management system across domains or sectors.

As with ISO/IEC 27001, controls are not applied wholesale. The standard defines a statement of applicability that documents the necessary controls and justifies each inclusion or exclusion (definition 3.26). An organisation may not need every Annex A control, and it can add controls of its own.

ISO 42001 and ISO 27001: what you can reuse

This is the most common question from organisations that already run an information security management system, and the answer is encouraging.

ISO/IEC 42001 states in its introduction that it applies the harmonized structure shared by ISO management system standards: identical clause numbers, clause titles, common text and core definitions. If you hold ISO/IEC 27001, you reuse the frame rather than rebuild it:

• context analysis and interested parties;
• control of documented information;
• internal audit programme and management review;
• nonconformity and corrective action;
• your risk assessment and treatment method, extended to AI risks.

What differs is the subject. ISO/IEC 27001 governs information security: confidentiality, integrity and availability. ISO/IEC 42001 governs how the organisation develops, provides or uses AI systems, and adds a requirement that ISO/IEC 27001 does not have: the AI system impact assessment (clauses 6.1.4 and 8.4), which looks at the consequences for individuals, groups and society that the system affects.

Put simply, ISO/IEC 27001 asks what can happen to your information. ISO/IEC 42001 also asks what can happen to people because of decisions made with AI. The two questions are different, so the two standards sit side by side and one does not replace the other. Organisations starting from scratch on both often build the ISO/IEC 27001 frame first, because customers ask for it more often, and add ISO/IEC 42001 on top.

ISO 42001 and the EU AI Act

This is where most commercial misunderstandings sit, so precision matters.

The AI Act is Regulation (EU) 2024/1689. Article 113 sets the calendar. The Regulation entered into force on 1 August 2024 and has applied in general since 2 August 2026, with exceptions: Chapters I and II, which include the prohibited practices and AI literacy, from 2 February 2025, with some prohibitions added in 2026 applying from 2 December 2026; governance and the rules on general-purpose AI models from 2 August 2025. Regulation (EU) 2026/1744, the Digital Omnibus on AI, in force since 27 July 2026, moved the high-risk requirements: they apply from 2 December 2027 for systems listed in Annex III and from 2 August 2028 for AI in products covered by Annex I legislation.

Presumption of conformity comes from the Official Journal

Under Article 40(1), high-risk AI systems and general-purpose AI models that conform to harmonised standards whose references have been published in the Official Journal of the European Union are presumed to conform to the requirements set out in Section 2 of Chapter III (Articles 8 to 15) or, for general-purpose AI models, to the obligations in Chapter V, Sections 2 and 3, to the extent those standards cover them. The presumption comes from that publication, not from how well known a standard is.

The AI Act harmonised standards are drafted by CEN and CENELEC, in the joint committee JTC 21, on the basis of a Commission standardisation request (Implementing Decision C(2025) 3871 of 23 June 2025). The first one is now referenced. Commission Implementing Decision (EU) 2026/2274 of 12 October 2026, published in the Official Journal on 14 October 2026 and in force from that day, publishes the reference of EN 18286:2026, Artificial intelligence, Quality management system for EU AI Act regulatory purposes. The decision states that the standard satisfies the requirements it aims to cover, which correspond to the request on quality management systems for providers of AI systems, including post-market monitoring.

ISO/IEC 42001 is not in that decision, and the European adoption EN ISO/IEC 42001:2026 does not change this: being published as a European standard is not the same as having a reference in the Official Journal under the AI Act. The Commission explains why in its FAQ on AI Act standardisation: "although ISO/IEC 42001:2023 helps to set up an AI management system, its goals and definitions are not aligned with the quality management system that is required under the AI Act". That is the reason it requested a new standard focused on regulatory compliance, which became EN 18286.

What Article 17 asks for

Article 17(1) requires providers of high-risk AI systems to put in place a documented quality management system, in written policies, procedures and instructions, covering at least a list of aspects. These include a strategy for regulatory compliance, design and development procedures, examination, test and validation, data management, the risk management system of Article 9, post-market monitoring, serious incident reporting and an accountability framework. Article 17(4) adds that, for financial institutions subject to internal governance rules under Union financial services law, most of this obligation is deemed fulfilled by complying with those rules.

The honest reading

• ISO/IEC 42001 certification does not make you AI Act compliant and does not give presumption of conformity.
• For providers of high-risk systems, the harmonised standard now referenced in the Official Journal for the quality management system is EN 18286:2026, and the obligations it addresses apply from 2 December 2027 or 2 August 2028.
• For everyone else, ISO/IEC 42001 builds the governance the Regulation takes for granted: an inventory of AI uses, roles, policies, risk and impact assessments, supplier controls and evidence. Many organisations are deployers rather than providers, and for them this governance is the practical starting point.

Where personal data are involved, the GDPR applies alongside all of this. We cover the data side in our guide to GDPR compliant AI.

How certification works

Certification follows the same model as other management system standards: an independent certification body audits the AI management system and issues the certificate. Three points are worth checking.

• Accreditation. In the EU, Regulation (EC) No 765/2008 gives each Member State a single national accreditation body, which accredits certification bodies. Management system certification bodies are accredited against ISO/IEC 17021-1, and for AI there is an additional standard, ISO/IEC 42006:2025, which sets requirements on top of ISO/IEC 17021-1 for bodies auditing and certifying an AI management system according to ISO/IEC 42001 (Standards Council of Canada catalogue). Ask for an accreditation whose scope covers ISO/IEC 42001 specifically.
• Separation of roles. ISO/IEC 17021-1 states that a certification body, and any part of the same legal entity, cannot offer or provide management system consultancy (clause 5.2.5). Whoever helps you build the system does not certify it.
• Scope. The certificate covers the scope you define in clause 4. A narrow scope is easier to certify and worth less in a customer questionnaire, so decide it with the questions you actually receive in mind.

The same checks we describe for choosing an ISO 27001 certification body apply here, and a body accredited for both standards can plan integrated audits.

When it makes sense to start

ISO/IEC 42001 is voluntary, so the useful question is what it unlocks. In the mid-market there are three common triggers.

1. A customer or tender asks how you govern AI. It is the same pattern that made ISO/IEC 27001 a de facto requirement in vendor assessments: first the question appears in the questionnaire, then it becomes a selection criterion.
2. An AI use that affects people. If a model contributes to decisions about candidates, customers or patients, the impact assessment the standard requires tells you first where you are exposed.
3. An existing compliance programme. If you already run ISO/IEC 27001, NIS2 or DORA work, you have the method, documentation and audit cadence. Adding the AI scope costs less than starting from zero closer to December 2027.

The sensible first step is not the certificate, it is the inventory: which AI systems are in use, in which processes, with which data and which vendors. That deliverable is useful with or without certification.

How we work on this at AmagisTech

We help mid-market organisations build the AI management system together with the information security one, instead of running two separate projects: mapping AI uses, risk and impact assessment, statement of applicability, continuous evidence collection and audit preparation. The scope and approach are on our ISO 42001 service page.

In short: ISO/IEC 42001 is the international standard for AI management systems, published in December 2023 and adopted in Europe as EN ISO/IEC 42001:2026. It shares the structure of ISO/IEC 27001, so it fits an existing management system, and adds the assessment of AI impacts on people. It is not a harmonised standard under the AI Act: since 14 October 2026 that role, for the quality management system of high-risk providers, belongs to EN 18286:2026.

If you want to understand where ISO/IEC 42001 fits in your compliance plan, talk to our team.