Frameworks / Global
ISO/IEC 42001
ISO/IEC 42001 is the first certifiable international standard for an AI management system (AIMS): the governance process an organisation operates to develop or use AI responsibly, and the only credential currently available to prove AI governance to a customer.
What it requires
Published in December 2023, ISO/IEC 42001 follows the same management-system structure as ISO 27001 (Clauses 4-10: context, leadership, planning, support, operation, performance evaluation, improvement), so an organisation with a working ISMS already knows the shape of the machine. Its distinctive requirements are in the AI domain: an AI policy and defined roles for AI governance; identification of the organisation's role across the AI lifecycle (developer, provider, user); AI risk assessment and treatment; and an AI system impact assessment, which considers consequences for individuals and society rather than only for the organisation. Annex A sets out controls covering areas such as AI policy, internal organisation, resources for AI systems, impact assessment, AI system lifecycle management, data for AI systems, information for interested parties, use of AI systems, and third-party and supplier relationships, with implementation guidance in Annex B. It is certifiable by accredited certification bodies on the familiar three-year cycle with surveillance audits. It is a management-system standard, not a technical safety benchmark: it does not tell you what accuracy or bias threshold your model must hit; it requires you to have decided, documented, justified and reviewed that.
ISO 42001 applies to any organisation that develops, provides or uses AI systems, at any size. Certification is voluntary. It is being pulled forward by three forces: enterprise customers whose procurement now includes AI questionnaires and who have no other credential to ask for; the EU AI Act, which creates governance and documentation obligations that an AIMS is well suited to satisfy; and boards asking who owns AI risk. For Israeli AI companies selling into European or US enterprises, it is rapidly becoming the AI equivalent of the ISO 27001 request: the thing an enterprise buyer names when they need proof rather than promises.
The controls people actually fail
- No inventory of AI systems in use: internal LLM tools, embedded model APIs and vendor AI features are adopted by teams with no central record, so the AIMS governs a subset of reality.
- AI system impact assessments are not performed, or are performed only for the flagship product and not for the models quietly making decisions inside HR, support or fraud workflows.
- Data provenance for training and fine-tuning is undocumented: nobody can state what data a model was trained on, under what lawful basis, or whether it may be used commercially.
- The organisation cannot articulate its role (developer, provider or user) for each AI system, which leaves the applicable controls and obligations undetermined.
- Human oversight is described in policy but not implemented: the decision is nominally reviewable but no human ever reviews it, and no evidence exists that anyone could.
- Third-party AI suppliers and model vendors are not assessed at all, despite carrying most of the actual risk.
- Model changes, retraining and prompt changes bypass change management entirely, so the system that was assessed is not the system in production.
What is ISO 42001 and can you be certified against it?
ISO/IEC 42001:2023 is the international standard specifying requirements for an AI management system (AIMS): the governance framework for developing, providing or using AI systems responsibly. Yes, it is certifiable: accredited certification bodies audit against it in a two-stage audit on a three-year cycle with annual surveillance, exactly as with ISO 27001. It is currently the only widely recognised third-party AI governance certification available.
Does ISO 42001 make us compliant with the EU AI Act?
No. ISO 42001 is a voluntary management-system standard; the EU AI Act is binding law with risk-tiered obligations, and certification against ISO 42001 does not by itself confer a presumption of conformity with it. What ISO 42001 does is build most of the machinery the AI Act expects (AI risk management, documentation, data governance, human oversight, lifecycle control and post-deployment monitoring), so a certified organisation is materially closer to AI Act readiness and can evidence its governance to customers today. Treat it as a strong foundation and an excellent commercial credential, not as a compliance shortcut.
How does ISO 42001 relate to ISO 27001?
They share the same management-system architecture (the Harmonized Structure, Clauses 4-10), which means an organisation with a working ISMS can extend it rather than start over: the same leadership commitment, risk process, internal audit, management review and improvement cycle carry across, and audits can often be combined. The difference is what they govern: ISO 27001 protects information and manages risk to the organisation; ISO 42001 governs AI systems and requires you to assess impacts on individuals and society, not only on yourself.
We only use AI vendors, we do not build models. Do we still need AI governance?
Yes, and this is the most common misunderstanding. ISO 42001 explicitly addresses organisations that use or provide AI systems, not just those that develop them. The risks of a purchased model are not smaller, merely less visible. A user organisation still needs an inventory of the AI systems in use, an understanding of what data flows into them, impact assessments for decisions those systems influence, human oversight where outcomes affect people, and assessment of the AI vendors themselves. In most mid-market companies, the AI actually in use was adopted by individual teams and never governed at all.