ISO 42001, published as ISO/IEC 42001:2023, is the first international standard for an AI management system. It sets out how to govern the AI you build or buy, through clauses 4 to 10 and 38 Annex A controls, with an AI system impact assessment at its core. It is certifiable, risk based, and not yet mandatory in Australia.
What is ISO 42001?
Most AI governance talk is slideware. ISO 42001 is the part an auditor can hold you to. Published by ISO and IEC in December 2023, ISO/IEC 42001:2023 is the world’s first certifiable management system standard for artificial intelligence. It does for AI what ISO 27001 does for information security: a documented, auditable system for governing AI across its life cycle, reviewed and improved over time, not a one off checklist. It applies whether you train your own models, embed someone else’s, or switch on AI features in tools you already run. An accredited body can certify you against it. It governs the system around the AI, not the model’s accuracy.
What does ISO 42001 require?
Two things: a management system, and a set of AI controls you justify. Clauses 4 to 10 carry the management system, context, leadership, planning, support, operation, performance evaluation and improvement, the same backbone as ISO 27001 and ISO 9001. Annex A then lists 38 controls across nine objectives. You do not apply all 38 by default. You select them against your AI risk and impact assessment and record every inclusion and exclusion, with reasons, in a Statement of Applicability. Annex B holds the implementation guidance for each control. The nine objectives:
| Annex A objective | What it governs |
|---|---|
| A.2 Policies related to AI | A documented AI policy, owned and reviewed |
| A.3 Internal organisation | Roles, responsibilities and reporting lines for AI |
| A.4 Resources for AI systems | The data, tooling, compute and people a system relies on |
| A.5 Assessing impacts of AI systems | Impact on individuals, groups and society |
| A.6 AI system life cycle | Responsible design, development, deployment and retirement |
| A.7 Data for AI systems | Provenance, quality and governance of training and operating data |
| A.8 Information for interested parties | Transparency to users, regulators and affected people |
| A.9 Use of AI systems | Responsible, intended and monitored use |
| A.10 Third party and customer relationships | Risk and accountability across the AI supply chain |
The selection and the Statement of Applicability are where readiness work concentrates. Our ISO 42001 readiness checklist walks the clauses and the controls in order.
What is an AI system impact assessment?
This is the part that is new. Information security risk asks what could happen to the organisation. ISO 42001 also asks what your AI could do to the people on the other side of it. The standard requires an AI system impact assessment in planning (clause 6) and its execution in operation (clause 8): a documented look at the potential consequences of an AI system for individuals, groups and society, covering fairness and bias, transparency, safety and accountability. Here is where teams trip. They run a security risk assessment, file it, and assume the impact assessment is done. It is a separate artefact, and an auditor will ask for it by name. We set out the method in AI risk assessment under ISO 42001.
Is ISO 42001 mandatory in Australia?
No. ISO 42001 is not mandated by Australian law, and there is no pass mark for an individual AI model. What has moved is the policy around it. In September 2024 the Department of Industry, Science and Resources published a Voluntary AI Safety Standard with 10 guardrails, and consulted on mandatory guardrails for AI in high risk settings. Those guardrails, accountability and governance, risk management, data governance, testing, human oversight, transparency, contestability, supply chain transparency, record keeping and stakeholder engagement, line up closely with ISO 42001. So an AI management system is a practical way to meet them ahead of any obligation. The pull today is commercial: customers, tenders and partners asking how you govern AI before they trust it with their data. Why AI governance matters now sets out that case.
How does ISO 42001 relate to the EU AI Act?
Certification is not compliance. The EU AI Act is binding law, in force from 1 August 2024 and phased in through 2025 to 2027. ISO 42001 is a voluntary standard. Certifying to ISO 42001 does not make you EU AI Act compliant, and the Act does not require the certificate. What the standard gives you is the governance scaffold the Act expects: risk management, data governance, human oversight, technical documentation and ongoing monitoring. Build the management system once and most of the Act’s process expectations are already in place. One is the law. The other is how you show you run a tidy shop. We compare them in full in ISO 42001 vs the EU AI Act.
How does ISO 42001 fit with ISO 27001?
It reuses the same backbone. Clauses 4 to 10 are shared with ISO 27001, so if you already run an information security management system, the context, leadership, internal audit and management review machinery extends to AI with far less duplication. The difference is scope. ISO 27001 protects information. ISO 42001 governs risks an ISMS never set out to address: bias, autonomy, explainability and the effect of an automated decision on a person. Run together, they share evidence and audit effort. If you are starting from an ISO 27001 base, the ISO 27001 guide shows how that management system is built.
How long does ISO 42001 take and what does it cost?
It depends on how many AI systems you run and how mature your governance already is. As an indicative guide from our own engagements, an AI governance gap assessment and AI system inventory takes 3 to 6 weeks, and full implementation to audit readiness 4 to 9 months. Pricing is not published, because it scales with the number and risk of your AI use cases, not a fixed rate. Certification, when you choose to pursue it, follows the ISO three year cycle: a Stage 1 documentation review, a Stage 2 implementation audit, then annual surveillance. We break down the drivers in ISO 42001 cost, and what it means for product teams in ISO 42001 for AI product companies.
Frequently asked questions
Yes. Accredited bodies certify an organisation’s AI management system on a three year cycle, like ISO 27001. There is no pass mark for an individual AI model; the certificate covers the governance system, not the model.
Annex A lists 38 controls across nine objectives, A.2 to A.10. You select them against your AI risk and impact assessment and justify each inclusion and exclusion in a Statement of Applicability.
No. The EU AI Act is binding law; ISO 42001 is a voluntary standard. Certification builds the governance the Act expects, but it does not replace legal compliance, and the Act does not require the certificate.
No, but it helps. The two standards share clauses 4 to 10, so an existing information security management system gives you most of the structure. ISO 42001 adds the AI specific controls and the AI system impact assessment.
Written by Gaurav Vikash, an ASD endorsed IRAP assessor and senior cyber security leader with 18 years of experience across Australia, the UK and Asia, including CISO and senior security leadership roles. He holds CISSP, CISA, CISM and CRISC and is an ISO 27001 and ISO 42001 Lead Implementer, and speaks regularly at industry conferences.
Sources:
- ISO/IEC 42001:2023, Information technology, Artificial intelligence, Management system, December 2023
- Voluntary AI Safety Standard, Department of Industry, Science and Resources, September 2024
- EU AI Act, European Commission, 2024
Last updated: 21 June, 2026