How it is structured
ISO lists ISO/IEC 42001:2023 as a published International Standard, Edition 1, from technical committee ISO/IEC JTC 1/SC 42, with the full title “Information technology — Artificial intelligence — Management system”. It is a management system standard: it asks an organisation to put policies and procedures in place for governing its use of AI, working through Plan-Do-Check-Act, rather than examining the detail of any one AI application.
ISO’s own summary lists the areas an AI management system has to cover:
- leadership and organisational context
- AI policy and objectives
- risk management for AI systems
- data governance and system lifecycle controls
- transparency and information provision
- performance evaluation and monitoring
- continual improvement
The standard text is a paid document; ISO’s page notes that customers with a billing address in the United Kingdom buy it from the BSI store. The public ISO pages we reviewed do not list the individual controls, so this guide does not reproduce clause or control wording. Work from the licensed standard when you map a client’s controls.
Which of your clients it applies to
ISO says the standard is for organisations of any size, in any industry, that develop, provide or use AI-based products or services, and that it covers public sector bodies and non-profits as well as companies. Its own examples include organisations that build AI systems, put AI into products or services, use AI for decisions or automation, or manage AI systems supplied by third parties.
ISO describes certification as voluntary, so whether a client needs it is usually a commercial question rather than a legal one. A customer questionnaire, a tender or an internal governance decision may raise it. A short triage works well on a first call: does the client build AI, embed it in what it sells, rely on it to make or support decisions, or buy AI from suppliers? A yes to any of these puts the client in ISO’s intended audience, whether or not it ever seeks a certificate.
How an MSP delivers it
The first six phases follow the order of ISO’s own list of practical first steps toward ISO/IEC 42001. The seventh is added because certification is optional and the client should choose it deliberately.
- Identify where AI is used. Build an inventory of AI systems the client develops, embeds or buys, including tools staff have adopted on their own.
- Define roles and responsibilities for AI oversight. Name who owns each system and who approves its use.
- Assess the risks. Record the risks of each AI system, such as bias, safety, security and misuse, and agree how they will be treated.
- Document policies. Write the AI use policy and the data governance rules the client will actually follow.
- Monitor performance and impacts. Decide what is measured for each system, how often, and who reviews it.
- Plan corrective actions and improvements. Keep a log of issues found and what was changed.
- Decide on certification. If the client wants independent confirmation, it engages an independent certification body; ISO notes these may be accredited by national accreditation bodies.
Agree the scope of the management system with the client in writing before phase one, so the inventory and the later audit cover the same boundary.
Requirements checklist
The first column uses the areas ISO lists in its public summary of the standard. The other two columns are working examples for an MSP, not text from ISO/IEC 42001.
| Area ISO lists | Question to ask the client | Evidence an assessor may ask to see |
|---|---|---|
| Leadership and organisational context | Who is accountable for AI use, and which interested parties care how it is used? | Named owner, scope statement, list of interested parties |
| AI policy and objectives | Is there a written AI policy, and does it set objectives that can be checked? | Approved policy, objectives with owners and review dates |
| Risk management for AI systems | How are risks assessed for each AI system, and who accepts them? | Risk register entries for AI systems, treatment decisions |
| Data governance and system lifecycle controls | Where does the data come from, and how are systems approved, changed and retired? | Data source records, change and release records |
| Transparency and information provision | What do users and affected people get told about the AI system? | User notices, supplier documentation, internal guidance |
| Performance evaluation and monitoring | How is performance checked after go-live, and what triggers a review? | Monitoring reports, review minutes, internal audit results |
| Continual improvement | What happened the last time something went wrong, and what changed? | Corrective action log, management review records |
Overlap with other frameworks
ISO/IEC 27001:2022 is Edition 3, published October 2022, titled “Information security, cybersecurity and privacy protection — Information security management systems — Requirements”. ISO describes a harmonized structure for its management system standards, so that users familiar with one find another intuitive, and sells ISO/IEC 42001 and ISO/IEC 27001 together as an AI and information security management package. The ISO pages we reviewed do not say clause by clause how the two align, so treat any reuse of a client’s existing ISMS (policy, internal audit, management review, corrective action) as something to confirm against both licensed texts. Our ISO 27001 guide covers the other standard.
ISO also states that ISO/IEC 42001 does not replace laws or regulations; it is a management framework that helps an organisation meet its obligations. We have not cited a published crosswalk between ISO/IEC 42001 and the EU AI Act or the NIST AI Risk Management Framework, so treat those as separate frameworks. The frameworks list shows both, and how control mapping works explains how overlaps are handled in practice. ISO’s related AI standards include ISO/IEC 22989 (terminology), ISO/IEC 23053 (a framework for AI systems that use machine learning) and ISO/IEC 23894 (guidance on AI risk management).
Common mistakes
- Treating it as a technical checklist. It is a management system standard. ISO says it works across the organisation rather than on the details of specific AI applications, so a model-by-model review does not replace policy, roles and review cycles.
- Assuming certification is compulsory. ISO describes it as voluntary. Ask what the client’s customers or tenders actually require before quoting for an audit.
- Assuming it settles legal compliance. ISO says it does not replace laws or regulations. Check separately which laws apply to the client’s AI use.
- Starting without an inventory. ISO’s first practical step is identifying where AI is used. Tools that staff adopt on their own are easy to miss.
- Quoting the standard loosely. Name the edition, ISO/IEC 42001:2023, when you tell a client what it is working toward, and confirm any certification scope with the certification body rather than assuming it.