What is ISO 42001?
ISO/IEC 42001:2023, published in December 2023, specifies requirements for an artificial intelligence management system: the set of policies, roles, processes and controls an organization uses to develop, provide or use AI systems responsibly. It applies whether you build models, ship products on top of someone else's, or only use AI inside the business. You define your role in the scope, and the controls you need follow from it.
The important word is management. The standard does not set accuracy thresholds or fairness metrics. It requires that you decide which ones matter for each system, measure them, act on the results and prove it to an auditor. A certified company can still ship a bad model. What it cannot easily do is ship one without anyone having assessed the risk, approved it and kept watching.
What the clauses require
Clauses 4 to 10 follow the harmonized structure every modern ISO management system standard uses, so anyone who has run ISO 27001 or 9001 will recognize the shape. The AI-specific content sits mostly in clause 6.
| Clause | What it asks for |
|---|---|
| 4 Context | Scope of the AI management system; your role as AI provider, producer, user or a mix; interested parties and their requirements |
| 5 Leadership | Top management commitment, an AI policy, assigned roles and authorities |
| 6 Planning | AI risk assessment, AI risk treatment with a statement of applicability, AI system impact assessment, measurable AI objectives |
| 7 Support | Resources, competence, awareness, communication, documented information |
| 8 Operation | Run the planned processes; repeat risk assessment, treatment and impact assessment at planned intervals and on significant change |
| 9 Performance evaluation | Monitoring and measurement, internal audit, management review |
| 10 Improvement | Nonconformity, corrective action, continual improvement |
The AI system impact assessment is the part with no ISO 27001 equivalent. Security risk assessment asks what can happen to the organization. Impact assessment asks what the AI system can do to people and groups outside it, such as an applicant scored unfairly or a patient given a wrong triage. ISO/IEC 42005:2025 gives detailed guidance on running one, and ISO/IEC 23894 covers AI risk management more broadly.
The Annex A controls
Annex A lists 38 reference controls grouped into nine objectives. As with ISO 27001, you are not required to implement all of them; you justify inclusions and exclusions in a statement of applicability. Annex B gives implementation guidance for each control.
| Control area | Covers |
|---|---|
| A.2 Policies related to AI | AI policy, alignment with other policies, review |
| A.3 Internal organization | Roles and responsibilities; how concerns are reported |
| A.4 Resources for AI systems | Documented data, tooling, compute, people for each system |
| A.5 Assessing impacts of AI systems | Impact assessment process on individuals, groups and society |
| A.6 AI system life cycle | Objectives, design, verification and validation, deployment, operation and monitoring, technical documentation, event logs |
| A.7 Data for AI systems | Data acquisition, quality, provenance and preparation |
| A.8 Information for interested parties | User documentation, incident communication, reporting obligations |
| A.9 Use of AI systems | Processes and objectives for responsible use; intended use |
| A.10 Third-party and customer relationships | Allocating responsibilities with suppliers and customers |
A.6 is where engineering carries the load: documented design decisions, verification and validation before release, monitoring in production, technical documentation and event logs. Teams that already generate model cards and evaluation reports from their pipelines pass this area with evidence they produce anyway. Teams that write documentation by hand before the audit tend to fail stage 2, because the records stop at the audit date.
How certification works
An accredited certification body runs a stage 1 audit (is the system designed and ready?) and a stage 2 audit (is it implemented and operating?). The certificate runs for three years with surveillance audits in years one and two and a recertification audit in year three. Since 2025, ISO/IEC 42006 sets extra competence requirements for bodies that audit AI management systems, which matters because the early market had certifiers with little AI depth. Check that the body is accredited by a recognized accreditation body for 42001 specifically, not only for 27001.
Adoption has been fast but is still small in absolute terms. The independent tracker iso42k.com counted 240 publicly announced certifications across 30 countries as of 8 November 2026, 129 of them in 2026. AWS, Anthropic and Microsoft (for Microsoft 365 Copilot) are among the companies that have announced certification.
ISO 42001 vs NIST AI RMF vs the EU AI Act
| ISO/IEC 42001 | NIST AI RMF | EU AI Act | |
|---|---|---|---|
| What is it? | Certifiable management system standard | Voluntary risk framework with a playbook | Binding regulation |
| Who checks? | Accredited certification body | You, or an assessor you hire | National market surveillance authorities; notified bodies for some systems |
| Output | Certificate, 3-year cycle with annual surveillance | Profiles and internal evidence | Conformity, CE marking, registration for high-risk systems |
| Best for | Proving governance to customers and auditors | Structuring an internal program fast, especially in the US | Mandatory if your AI reaches the EU market |
They are complements, not alternatives. The NIST AI RMF is the faster way to structure risk work, and NIST's AI Resource Center publishes crosswalks between the RMF and ISO AI standards, so most work done under one counts toward the other. ISO 42001 supports EU AI Act readiness, especially the quality management and risk management obligations, but it is not a harmonised standard under the Act and gives no legal presumption of conformity on its own.
When certification pays off
- You sell AI to enterprises. Security reviews now ask for it, and a certificate replaces dozens of questionnaire answers. This is the clearest return.
- You already run ISO 27001. The marginal cost is the AI-specific work plus extra audit days, not a new management system.
- You need a governance backbone before EU high-risk obligations land in December 2027. The management system gives the program a structure that survives reorgs.
If none of those apply, adopt the standard internally, run the impact assessments and controls that matter for your systems, and wait until a customer or regulator asks for the certificate. The framework is most of the value. The certificate is a sales and assurance asset.
Related questions
Frequently asked questions
What is ISO 42001?
ISO/IEC 42001:2023 is the international standard for an artificial intelligence management system (AIMS), published in December 2023 by ISO and IEC. It sets requirements for establishing, running and continually improving the policies, roles, risk and impact assessments and controls an organization uses to develop, provide or use AI responsibly. It is the first AI standard organizations can be certified against.
How do you get ISO 42001 certified?
Define the scope, run AI risk and impact assessments, select and implement controls (Annex A gives the reference set) and document them in a statement of applicability, then operate the system long enough to produce records, run an internal audit and a management review. An accredited certification body then runs a stage 1 audit of design and readiness and a stage 2 audit of implementation. Certificates last three years with annual surveillance audits.
How long does ISO 42001 certification take?
Organizations that already hold ISO 27001 commonly report three to six months from kickoff to certificate, because the management system scaffolding exists. Starting from scratch, six to twelve months is more typical. The limiting factor is usually not documentation but having enough operating history, such as completed risk assessments, internal audit and management review records, for the stage 2 auditor to test.
How much does ISO 42001 certification cost?
Audit fees are set from audit-day tables that scale with headcount, scope and complexity, so there is no list price. Published industry estimates for audit plus preparation commonly run from the low tens of thousands of dollars for a small, narrowly scoped company to well into six figures for a large enterprise once internal staff time and consultants are included. Combining the audit with an existing ISO 27001 audit is the most reliable way to cut cost.
Which companies are ISO 42001 certified?
Public announcements include AWS, Anthropic and Microsoft (for Microsoft 365 Copilot), along with many AI start-ups and SaaS companies. The independent tracker iso42k.com counted 240 publicly announced certifications across 30 countries as of 8 November 2026, 129 of them in 2026, and notes the real number is higher because many organizations never announce. Schellman, A-LIGN, SGS and BSI were the most common certifiers in its data.
Does ISO 42001 replace ISO 27001?
No. ISO 27001 covers information security; ISO 42001 covers how AI is governed, including impacts on people that are not security issues, such as bias and transparency. They share the same high-level structure, so most organizations run them as one integrated management system with shared document control, internal audit and management review.
Sources
- ISO, ISO/IEC 42001:2023 Information technology: Artificial intelligence: Management system; ISO, ISO/IEC 42006:2025.
- iso42k.com, State of ISO 42001 certification (data through 8 November 2026).
- Anthropic, Anthropic achieves ISO 42001 certification for responsible AI; Microsoft, Microsoft 365 Copilot and ISO 42001.