ISO 42001: the AI management system standard

ISO 42001 is the international standard that sets the requirements for an artificial intelligence management system. Published in December 2023 by ISO and IEC, it applies to any organisation that develops, provides or uses AI systems, and it can be certified.

For a CISO or compliance manager, the good news is that it looks a lot like ISO 27001: same frame, same continual improvement logic. This guide explains what it contains, what it adds to an existing information security management system and how to prepare.

ISO/IEC 42001:2023 specifies requirements for establishing, implementing, maintaining and continually improving an artificial intelligence management system (AIMS). It is meant to help organisations develop, provide or use AI systems responsibly, in line with applicable requirements and the expectations of interested parties, and it applies to organisations of any size, type and nature.

Like any management system standard, it does not cover a specific product or algorithm but how the organisation governs its use of AI: who decides, on which criteria, with which controls and what evidence. ISO describes it as the world's first standard of its kind and notes that it follows the Plan-Do-Check-Act approach.

AI systems raise risks that traditional security frameworks cover poorly: bias, lack of transparency, misuse, dependence on training data, effects on people who do not know an AI is involved. ISO's overview mentions bias, safety, security and misuse in particular.

There is also an organisational problem: AI often arrives through business units, via online services or features built into software, without going through IT. This is shadow AI. A management standard forces you to inventory these uses, assign responsibilities and decide explicitly what is acceptable. As ISO points out, it does not replace the law, but it provides the organisation needed to track applicable requirements, including those of the EU AI Act, covered in a dedicated series.

ISO 42001 follows the harmonized structure shared by ISO management system standards: the same clause titles, some identical text and shared definitions, so an organisation can run several standards in one system. In practice:

• Clauses 4 to 10, mandatory: context of the organisation, leadership, planning, support, operation, performance evaluation, improvement.
• Annexes: Annex A provides reference control objectives and controls for AI systems, against which risk treatment is compared, supplemented by implementation guidance and examples of objectives and risk sources.

The heart of the standard is in the planning clause. It requires three documented exercises:

• an AI risk assessment: identify, analyse and evaluate risks against defined criteria, as in ISO 27005;
• treatment of those risks, with controls compared against Annex A and a statement of applicability;
• an AI system impact assessment covering effects on individuals, groups and society, including misuse.

These are not one-off exercises: the operation clause requires them to be repeated at planned intervals and after significant changes (new model, new data, new use). Two companion standards help: ISO/IEC 23894 (2023) gives guidance on AI risk management, and ISO/IEC 42005 (May 2025) describes how to conduct an AI system impact assessment. The logic is similar to the GDPR's data protection impact assessment (DPIA), with a wider scope than personal data alone.

Because both standards share the same structure, an organisation with an ISO 27001 ISMS does not have to start from scratch. The diagram below shows, clause by clause, what already exists and what ISO 42001 adds. Pick a clause to see the detail.

The division of roles remains clear: the security of AI systems (protecting models, training data and access) still belongs to the ISMS and the ISO 27001 Annex A. ISO 42001 covers what security does not: the purpose of systems, their effects on people, data quality, transparency and human oversight. One security policy plus a separate AI policy, one internal audit cycle and a joint management review are usually enough to steer both.

Certification is voluntary. As ISO notes, ISO does not certify organisations: independent certification bodies do, and they may be accredited by a national accreditation body. Since July 2025, ISO/IEC 42006 sets the additional requirements for bodies that audit and certify AI management systems, on top of the general rules for management system certification.

The process mirrors other management system standards: a two-stage initial audit, then surveillance audits and recertification. Auditors will focus on the inventory of AI systems, the AI policy, risk and impact assessments, the statement of applicability and evidence that they are kept up to date.

1. Inventory AI systems developed, bought or used, including AI features built into everyday software, and record the organisation's role for each (provider, user…).
2. Set the scope of the management system: the whole organisation, or one product or activity first.
3. Write the AI policy and have management approve it, and appoint owners.
4. Assess risks and impacts on priority systems, with written criteria.
5. Lean on the ISMS for everything shared: document control, internal audits, management review, assessment of suppliers, especially those providing AI models or services.

As with any new framework, the challenge is to avoid a second pile of documents next to the first. Linking ISO 42001 requirements to ISO 27001 and the other frameworks you follow means assessing shared items once, within a single governance, risk and compliance approach. And if AI is also entering your own compliance tools, our article on AI and GRC covers what it brings and its limits.

Going further: Phinasoft's policies editor handles any standard and lets you link frameworks together.