TL;DR
What is ISO 42001? It is the world’s first certifiable standard for AI management systems, published December 18, 2023. It applies to any firm that develops, provides, or uses AI, and it shares its structure with ISO 27001 and ISO 9001.
- Seven requirement clauses, Clauses 4 through 10, built on the same PDCA cycle as ISO 27001.
- 38 Annex A controls across 9 themes, plus four informative annexes for guidance.
- Two required assessments: an AI risk assessment (Clause 6.1.2) and an AI system impact assessment (Clause 6.1.4). They are not the same document.
- Certification runs a two-stage audit, then annual surveillance, then full recertification every three years.
- AWS, Google, Microsoft, Anthropic, and CrowdStrike have all certified since late 2024.
Sources: ISO/IEC 42001:2023 · Regulation (EU) 2026/1744
What ISO/IEC 42001 is #
ISO/IEC 42001:2023 gives you a structured way to govern how your firm builds, deploys, and uses AI. ISO/IEC JTC 1/SC 42 published it on December 18, 2023. It is the first standard in the ISO family built for AI. It is also the first one you can certify against.
If you know ISO 27001 or ISO 9001, this will feel familiar. All three share the Annex SL high level structure, the common skeleton ISO uses across its management system standards. That shared skeleton is deliberate. A firm with ISO 27001 already in place can extend its existing management system into ISO 42001 rather than start from zero.
The substance still differs. ISO 27001 protects your information. ISO 42001 governs a system that can learn, drift, and produce a biased or opaque result you did not design for. That difference shapes every clause below.
What the standard requires #
ISO 42001 runs on the same Plan Do Check Act cycle as its ISO siblings, spread across seven requirement clauses.
| Clause | Title | What it requires |
|---|---|---|
| 4 | Context of the organization | Define your AIMS scope. Name the internal and external issues, and the interested parties, that bear on your AI use. |
| 5 | Leadership | Top management commits. You write an AI policy and assign roles and authority. |
| 6 | Planning | Run an AI risk assessment (6.1.2) and an AI system impact assessment (6.1.4). Set measurable AI objectives. Produce your Statement of Applicability. |
| 7 | Support | Build competence and awareness. Set up communication. Keep documented information. |
| 8 | Operation | Run your risk treatment plan. Operate controls across the AI system lifecycle. |
| 9 | Performance evaluation | Monitor and measure the AIMS. Run internal audits. Hold management reviews. |
| 10 | Improvement | Fix nonconformities. Keep improving the AIMS. |
Clauses 6.1.2 and 6.1.4 do the most work, and they are easy to conflate. Our guide to the AI risk assessment against the AI system impact assessment covers this. It explains why auditors treat them as two separate documents, not one.
What the Annex A controls cover #
Four annexes back the seven clauses, all informative rather than mandatory. Annex A holds 38 controls across 9 themes. Annex B gives implementation guidance for each one. Annex C is a starting catalogue of AI risk sources for your own risk assessment. Annex D covers applying the AIMS across different sectors.
| Theme | Covers |
|---|---|
| A.2 Policies related to AI | Your documented commitments for responsible AI development and use. |
| A.3 Internal organization | Roles, responsibility, and accountability for AI governance. |
| A.4 Resources for AI systems | Competence, training, and infrastructure. |
| A.5 Assessing impacts of AI systems | The impact assessment process itself. |
| A.6 AI system life cycle | Design, development, testing, deployment, monitoring, retirement. |
| A.7 Data for AI systems | Data quality, provenance, and bias in training data. |
| A.8 Information for interested parties | What you disclose to the people your AI affects. |
| A.9 Use of AI systems | Acceptable use, human oversight, monitoring in operation. |
| A.10 Third party and customer relationships | Governance of AI you buy, sell, or build on someone else’s platform. |
Our guide to ISO 42001 clauses and Annex A controls has the full clause by clause breakdown. It also covers where organizations get the theme count wrong.
Who should get certified #
The short answer covers almost everyone: any firm that develops, provides, or uses AI, regardless of size or sector. The more useful answer is about where the standard pays off fastest.
| Organization type | Why it matters here |
|---|---|
| AI product and platform firms | Shows customers and regulators you build responsibly, in a market where trust is a selling point. |
| Enterprises deploying AI internally | Governs AI used in HR, finance, and operations, where an ungoverned tool creates real liability. |
| Consulting and professional services firms | Backs up your AI advisory work with a credential your clients can check. |
| Healthcare organizations | Covers AI used in diagnostics and triage, alongside your existing health privacy duties. |
| Financial services firms | Covers AI in credit scoring and fraud detection, where the regulatory and ethical exposure is high. |
| Public sector agencies | Backs citizen facing AI, such as benefits decisions, where fairness has to be demonstrable. |
What is driving adoption #
Regulation #
The EU AI Act took legal effect on August 1, 2024, with its obligations phased in over several years. Its high risk system rules map to ISO 42001’s own control set: risk management, transparency, documentation, human oversight. ISO 42001 is not a harmonized standard under the Act, so certification alone does not clear your Act obligations. It still builds most of the documentation a high risk deployer needs.
The Act’s own timeline moved this year. A Digital Omnibus agreement pushed the deadline for standalone high risk systems from August 2026 to December 2, 2027. High risk systems built into already regulated products now have until August 2, 2028. Regulation (EU) 2026/1744 made both dates binding when it took effect on July 27, 2026.
Canada has no equivalent federal statute. The proposed Artificial Intelligence and Data Act died on prorogation in January 2025 and has not returned. A firm operating in both markets has one binding framework to track: the EU’s. It has one voluntary standard to build its AI governance around in the meantime: ISO 42001.
Customer demand #
A-LIGN’s 2026 compliance benchmark report found that 80% of companies using AI are already fielding customer questions about their AI risk management. The same report found a third of companies have no AI compliance strategy at all. That gap is where a certified AIMS earns its keep, the same way ISO 27001 became a standard procurement ask for information security.
Risk and liability #
A biased hiring model, a chatbot that fabricates answers, an opaque credit denial. Each one generates lawsuits, regulatory attention, and a damaged reputation. Certification does not remove that risk. It gives you a documented, auditable process for finding and treating it before it becomes a headline.
Who has already certified #
Certification moved from a novelty to a norm inside two years.
| Organization | Certified | Certification body |
|---|---|---|
| Synthesia | September 2024 | A-LIGN |
| KPMG Australia | October 2024 | BSI |
| AWS | November 2024 | Schellman |
| December 2024 | — | |
| Anthropic | January 2025 | Schellman |
| Microsoft | 2025 | — |
| CrowdStrike | January 2026 | — |
How to implement an AIMS #
Implementation runs the same PDCA cycle the standard is built on.
- Secure leadership commitment. Clause 5 needs a named owner with real budget and authority, not a policy nobody funded.
- Define your scope. Clause 4 has you name which AI systems, teams, and locations the AIMS covers.
- Run a gap analysis. Check your current practice against all seven clauses and the Annex A themes. Existing ISO 27001 work covers more of this than you expect.
- Run both assessments. The AI risk assessment and the AI system impact assessment are separate documents with separate methods. See our guides to conducting the risk assessment and to how the two differ.
- Write your AI policy and Statement of Applicability. The policy sets your commitments. The SoA maps every Annex A control to your context, included or excluded, with a reason either way. Our guide to writing an AIMS policy covers the first part.
- Implement the controls, then audit. Work through the nine Annex A themes and keep the records Clause 7.5 requires. Then run your internal audit and management review, before the certification body arrives for Stage 1 and Stage 2.
Most firms take three to six months. A firm with ISO 27001 already running moves faster, since the internal audit program and management review already exist.
Frequently asked questions #
What is ISO/IEC 42001:2023?
It is the first international standard for AI management systems, published December 18, 2023 by ISO/IEC JTC 1/SC 42. It sets requirements for building, running, and improving a management system for any firm that develops, provides, or uses AI. It follows the same Plan Do Check Act structure as ISO 27001 and ISO 9001.
Who should get ISO 42001 certified?
Any firm that develops, provides, or uses AI, regardless of size or sector. AI product firms, enterprises running AI in-house, and consulting firms giving AI advice all see real value from certification. So do healthcare, financial, and public sector organizations.
How does ISO 42001 relate to the EU AI Act?
The two overlap a great deal but are not the same thing. ISO 42001 is not a harmonized standard under the Act, so certification alone does not clear your legal obligations. It still builds most of the documentation a high risk deployer needs. The Act’s own high risk deadline moved to December 2027 for standalone systems, under a 2026 rule change. That eases the pressure to align the two this year, though it is still worth doing.
What are the main benefits of certification?
You get a documented process for finding and treating AI specific risks, plus independent proof of that process from an accredited auditor. You also get less duplicated work if you already hold ISO 27001 or ISO 9001. More customers now ask AI vendors for this kind of proof.
How long does implementation take?
Three to six months for most firms. The exact time depends on how many AI systems are in scope and how mature your existing management system already is. A firm with ISO 27001 in place moves toward the faster end.
Does Canada have its own version of ISO 42001?
No. Canada’s proposed Artificial Intelligence and Data Act died on prorogation in January 2025 and has not been reintroduced. Canada has no binding federal AI statute. That gap makes a voluntary standard like ISO 42001 the main structured option for a Canadian firm’s AI governance.
Key takeaways #
- ISO 42001 is the first certifiable AI management system standard, published December 2023.
- It shares its PDCA structure with ISO 27001 and ISO 9001, so existing certifications shorten the path.
- 38 Annex A controls across 9 themes, not the 6 themes some guides still cite.
- Clause 6.1.2’s risk assessment and Clause 6.1.4’s impact assessment are two documents, not one.
- The EU AI Act’s high risk deadline moved to December 2027, so ISO 42001 remains the faster path to structured governance right now.
- Canada has no equivalent federal statute. AIDA died on prorogation and has not returned.
- AWS, Google, Microsoft, Anthropic, and CrowdStrike have all certified since late 2024.
- Most implementations take three to six months.
Build an AIMS with a compliance manager who has done it before #
Nank.ai runs ISO 42001 programmes for companies in Canada and the United States, from the first gap analysis through Stage 2, with a dedicated compliance manager who owns the result.
