ISO 42001 has the same clauses as ISO 9001 and ISO 27001, plus an assessment that those standards do not have: what your AI means for people. What do you really have to do of that, and where does it differ from standards you may already know? After reading this article you will know how the requirements are structured, what each clause requires, what the difference is between the risk assessment and the impact assessment, what is in annex A, which controls you may drop and which edition applies.
Structure of the ISO 42001 requirements
ISO 42001 has two kinds of requirements. Clauses 4 to 10 describe your management system for AI. Those requirements always apply. Annex A gives control objectives and controls, and annex B explains how to implement them. Both annexes are normative. So the auditor can assess you against them. Annexes C and D are informative. They help you, but you are not assessed against them.
| Part | What it contains |
|---|---|
| Clauses 1 to 3 | the scope of the standard, references and terms |
| Clauses 4 to 10 | the requirements for your management system for AI |
| Annex A (normative) | control objectives and controls as a reference list |
| Annex B (normative) | guidance on implementing those controls |
| Annex C (informative) | possible objectives for AI and sources of risk |
| Annex D (informative) | the use of the system in different sectors |
Clauses 4 to 10 follow the fixed structure that ISO uses for all its management system standards (harmonized structure). ISO 9001 and ISO 27001 have the same numbers and titles. The standard says in its introduction that you integrate your management system for AI into your existing processes and your existing management. So it is not a separate project.
Clause 4 of ISO 42001: context of your organisation
You determine which issues inside and outside your company affect your AI, who has an interest in it and what they expect. Then you record the scope (4.3): which AI systems, which role and which locations are covered. A company that builds AI has a different scope from a company that buys in an AI tool.
Clause 5 of ISO 42001: leadership
Top management carries the system. It sets an AI policy (5.2) and assigns roles and authorities (5.3). The policy fits the purpose of your company and provides the framework for your AI objectives.
Clause 6 of ISO 42001: planning
Clause 6.1 has three parts that make ISO 42001 distinctive: the AI risk assessment (6.1.2), the AI risk treatment (6.1.3) and the AI system impact assessment (6.1.4). In addition, you set objectives for your AI and plan how to achieve them (6.2). You plan changes to your system in advance (6.3).
Clause 7 of ISO 42001: support
You provide resources, people who are competent for their task and awareness (7.1 to 7.3). You arrange communication (7.4) and you control what you record (7.5). Anyone who works on your AI must know what your AI policy requires of them.
Clause 8 of ISO 42001: operation
You carry out what you planned in clause 6. You control your work (8.1) and actually carry out the risk assessment (8.2), the risk treatment (8.3) and the impact assessment (8.4). Plan when you repeat those assessments, for example when an AI system changes because of a new model or new data.
Clause 9 of ISO 42001: performance evaluation
You measure whether your system works (9.1). You carry out an internal audit (9.2) and a management review (9.3). The latter is a fixed moment at which management looks at whether the system works.
Clause 10 of ISO 42001: improvement
You continually improve your system (10.1). When something goes wrong, you look for the cause and take a measure that prevents recurrence (10.2).
Risk assessment and impact assessment under ISO 42001
The risk assessment asks what can go wrong with your AI and what the consequences are. The impact assessment asks, for each AI system, what it can mean for the people and groups who deal with it. Take a recruitment agency that has an AI tool pre-select job applicants. The risk for the agency is that the tool misses good candidates. The impact on people is that one group of applicants systematically scores lower. Both belong in your system.
The introduction of the standard explains why AI needs specific attention:
- Decisions. AI sometimes makes decisions automatically, in a way that cannot always be followed or explained.
- Data. AI systems are built with data and machine learning, and less with rules written by people.
- Learning. A system that keeps learning changes its behaviour while you use it.
The introduction also names topics for which you need processes: safety, fairness, transparency, the quality of data and the quality of AI systems, throughout their entire life cycle. Guidance on the impact assessment is given in the separate standard ISO/IEC 42005:2025. That standard helps you find and record the consequences for individuals, groups and society. It is a guide. You cannot be certified against it.
Annex A of ISO 42001: control objectives and controls
Annex A is a reference list of control objectives and controls for AI. Annex B gives guidance on implementing each control. You choose controls on the basis of your risk assessment and your impact assessment. The introduction of the standard says that you may apply the requirements with a risk-based approach. That way the level of control fits each application of AI in your company. A company that only uses AI chooses different controls from a company that trains its own models.
Excluding a control from annex A of ISO 42001
You do not exclude a requirement from clauses 4 to 10. They apply to every company, of every size. You may, however, drop a control from annex A if your risks do not make that control necessary. You then record which controls you apply and why you exclude others, in a Statement of Applicability. That requirement is in clause 6.1.3. It works the same way for ISO 27001. Each row in that Statement names a control, whether you apply it and why. Ask your certification body before stage 1 how it wants to see your Statement.
Take an online shop that only uses a bought-in chatbot. If a control is about developing your own AI, it is usually not needed there. The agreements with the supplier of the chatbot are then all the more important. The auditor may record an exclusion without a reason as a nonconformity.
Applicable edition of ISO 42001
The applicable edition is ISO/IEC 42001:2023. That is the first edition, from December 2023. There is no amendment and no transition under way. In March 2026 CEN adopted the standard without changes as EN ISO/IEC 42001:2026. In the Netherlands it is called NEN-EN-ISO/IEC 42001:2026. That version replaced NEN-ISO/IEC 42001:2025 on 1 March 2026. The text is the same as that of ISO.
ISO 42001 requirements and the choice of your certification body
The requirements are the same at every certification body. What differs is how well the auditor knows your kind of AI. ISO/IEC 42006:2025 is the standard with requirements for certification bodies of ISO 42001. That standard sets general and specific technical requirements for the knowledge of the people who audit. So in the quote, ask what experience the auditor has with your AI: developing it yourself, providing it or using it. Also ask about experience in your sector. Put those questions in one request through CertificeerWijzer to certification bodies with ISO 42001 in their accreditation. Then you put their answers about the auditor side by side. What you need to record is in the documents for ISO 42001.
Sources
- EN ISO/IEC 42001:2026, preview copy with European foreword, table of contents, introduction and scope (via SIST) cdn.standards.iteh.ai
- ISO, ISO/IEC 42001:2023 iso.org
- NEN, NEN-EN-ISO/IEC 42001:2026 en nen.nl
- ISO/TMBG, Harmonized structure for MSS with guidance for use (2026) iso.org
- ISO/IEC 42006:2025, preview copy with table of contents (via SIST) cdn.standards.iteh.ai