Key takeaways
- ISO/IEC 42001:2023 is the first certifiable standard for an AI management system (AIMS). ISO and IEC published it in December 2023, developed by the committee ISO/IEC JTC 1/SC 42. A certificate is valid for three years, with annual surveillance audits.
- The standard follows the same structure as ISO 27001: Clauses 4 to 10 set the management system requirements, Annex A contains 38 controls in nine areas (A.2 to A.10), and Annexes B to D provide implementation guidance.
- ISO 42001 is voluntary and is not a harmonized standard under the EU AI Act. A certificate therefore proves neither conformity with the law nor that an AI system can withstand attacks.
- Compared with an ISMS under ISO 27001, the main additions are an AI policy, an AI inventory, the AI system impact assessment (Clauses 6.1.4 and 8.4), and controls for data, the AI life cycle, and AI suppliers. You reuse the framework, risk method, and audit process.
- With an existing ISMS, ISO 42001 certification is realistic in as little as 1 to 3 months; without one, in as little as 4 to 6 months. These are lower bounds, not averages.
Last updated: November 5, 2026
ISO 42001 is the international standard for managing artificial intelligence responsibly within a company, and the only one against which you can certify an AI management system. It sets out how you record AI systems, assess risks and impacts, assign responsibilities, and keep improving the whole setup. That applies whether you develop AI yourself, build it into your product, or simply use it as a tool.
Demand is growing faster than the number of certificates. Procurement teams increasingly ask for proof of AI governance in supplier assessments. At the same time, ISO's annual survey counted just under 300 valid ISO/IEC 42001 certificates worldwide at the end of 2025, as the ISO survey shows. A certificate is still a real differentiator.
This guide explains what the standard contains, who it is worth it for, what you can reuse from an existing ISMS, and how certification works. Kertos has held ISO 42001 certification itself since 2025, and where it helps, we show how our own audit went.
What is ISO/IEC 42001?
ISO/IEC 42001 is a management system standard that sets requirements for an AI management system (AIMS). An AIMS is the combination of policies, roles, processes, and evidence a company uses to govern how it develops, provides, and uses AI. The standard applies to companies of any size and sector that provide or use AI systems.
ISO and IEC published the standard jointly, and subcommittee ISO/IEC JTC 1/SC 42 on artificial intelligence developed it. The current and so far only edition is ISO/IEC 42001:2023 from December 2023, as the official ISO entry shows. Europe has adopted it as EN ISO/IEC 42001.
| Attribute | ISO/IEC 42001 |
|---|---|
| Full title | Information technology, Artificial intelligence, Management system |
| Current version | ISO/IEC 42001:2023, edition 1 |
| Published | December 2023 |
| Committee | ISO/IEC JTC 1/SC 42 (Artificial intelligence) |
| European adoption | EN ISO/IEC 42001 |
| Type | Certifiable management system standard |
| Structure | Clauses 4 to 10 (requirements), Annex A (38 controls), Annexes B to D (guidance) |
| Certificate validity | Three years, with annual surveillance audits |
| Requirements for certification bodies | ISO/IEC 42006:2025 |
| Related standards | ISO/IEC 27001 (ISMS), ISO/IEC 42005:2025 (AI system impact assessment), ISO/IEC 23894:2023 (AI risk management), ISO/IEC 27090:2026 (AI security) |
A common misconception: ISO 42001 checks whether an individual AI system is fair, secure, or accurate. The auditor examines the management system, meaning whether you systematically identify, assess, treat, and monitor the risks and impacts of your AI systems. A certificate attests to governance, not to the quality of a specific model.
Structure of the standard: Clauses 4 to 10 and Annexes A to D
ISO 42001 follows the Harmonized Structure that ISO sets for all management system standards. Clauses 1 to 3 cover scope, references, and terms, and Clauses 4 to 10 contain the requirements the auditor checks. If you know ISO 27001, you will find the same headings and often the same core text.
Watch the terminology here: the Harmonized Structure has nothing to do with a harmonized standard under EU law. It only describes the common layout of ISO management systems. The following table shows where ISO 42001 becomes AI-specific in each clause.
| Clause | Topic | What ISO 42001 adds for AI |
|---|---|---|
| 4 Context of the organization | Internal and external issues, interested parties, scope | Determine your role, such as AI provider, AI producer, or AI user (4.1), and define the AI systems in scope (4.3) |
| 5 Leadership | Management responsibility, policy, roles | An AI policy that fits the company's purpose (5.2) and roles for the AIMS (5.3) |
| 6 Planning | Risks, opportunities, objectives | AI risk assessment (6.1.2), risk treatment with a Statement of Applicability (6.1.3), AI system impact assessment (6.1.4), AI objectives (6.2) |
| 7 Support | Resources, competence, awareness, communication, documentation | Demonstrate competence and awareness on AI topics (7.2, 7.3) |
| 8 Operation | Operational planning and control | Perform risk assessment, risk treatment, and impact assessment regularly and after significant changes (8.2 to 8.4) |
| 9 Performance evaluation | Monitoring, internal audit, management review | Internal audit (9.2) and management review (9.3) for the AIMS |
| 10 Improvement | Nonconformities, corrective action, continual improvement | Same logic as ISO 27001 |
Four annexes complete the standard. Annex A contains the 38 controls from which you derive your Statement of Applicability. Annex B gives implementation guidance for each of these controls, much as ISO 27002 does for ISO 27001. Annex C lists possible AI-related objectives such as fairness, transparency, robustness, and privacy, along with typical risk sources such as the level of automation, lack of explainability, or the technology readiness of a method. Annex D describes how to use the AIMS across sectors and integrate it with other management systems.
Annex A controls at a glance
Annex A contains 38 controls in nine areas, numbered A.2 to A.10. As with ISO 27001, you select controls based on your risk assessment and justify in the Statement of Applicability which ones you implement and which you exclude (Clause 6.1.3). Annex A is not exhaustive, so you can add controls from other sources.
| Area | Controls | What it covers | Example of evidence |
|---|---|---|---|
| A.2 Policies related to AI | 3 | AI policy, alignment with other policies, regular review | Approved AI policy with a review date |
| A.3 Internal organization | 2 | Roles and responsibilities, reporting of concerns about AI | Role descriptions, a documented reporting channel |
| A.4 Resources for AI systems | 5 | Documentation of data, tooling, system, computing, and human resources | AI inventory listing models, data sources, infrastructure, and owners |
| A.5 Assessing impacts of AI systems | 4 | Process and documentation, impacts on individuals, groups, and society | Impact assessment per AI system with outcome and approval |
| A.6 AI system life cycle | 9 | Objectives and processes for responsible development, requirements, design, verification and validation, deployment, operation and monitoring, technical documentation, event logs | Test records before release, monitoring reports |
| A.7 Data for AI systems | 5 | Data management, acquisition, data quality, provenance, preparation | Data sheet with the provenance and quality criteria of training data |
| A.8 Information for interested parties | 4 | Documentation for users, channels for reporting adverse impacts, incident communication, information obligations | User notices, a reporting form, an incident communication template |
| A.9 Use of AI systems | 3 | Processes and objectives for responsible use, intended use | Approval process for new AI tools, an acceptable use policy |
| A.10 Third-party and customer relationships | 3 | Allocation of responsibilities, suppliers, customers | Supplier assessment for AI vendors, contractual terms |
Which controls apply depends on your role. If you build AI into your own products, you mainly need the areas covering the life cycle, data, and information for users (A.6 to A.8). If you only use AI, you focus on responsible use and suppliers (A.9 and A.10) and can exclude development-only controls with a justification.
Compared with the 93 controls in ISO 27001 Annex A, the catalog looks lean. The effort, however, sits less in the number of controls than in the processes behind them, especially the inventory and the impact assessment. Our overview of the ISO 27001 Annex A controls shows how control selection works for ISO 27001.
Who is ISO 42001 worth it for?
ISO 42001 pays off most for companies that build AI into their products and sell them to other businesses. For them, a certificate replaces many individual questions in tenders and supplier questionnaires. Typical candidates are:
- Providers of AI products and B2B SaaS with AI features whose enterprise customers require proof of AI governance.
- Suppliers that receive questionnaires about AI use from their customers, for example in the automotive or financial sector.
- Companies in regulated sectors such as healthcare or finance that use AI for decisions about people.
- Companies with several AI use cases across teams that need one consistent process from idea to operation.
If you only use a single AI tool internally, you usually do not need a certificate. An AI policy, training, and an AI inventory are enough to meet your legal obligations. Our guide How to prepare your business for the EU AI Act explains how to go about it.
ISO 42001 and the EU AI Act
Many teams assume the EU AI Act makes ISO 42001 mandatory. It does not. ISO/IEC 42001 is not a harmonized standard, so a certificate does not trigger a presumption of conformity with the law. For the quality management system that providers of high-risk AI need under Art. 17 AI Act, however, EN 18286 has been published in the Official Journal as a harmonized standard since October 14, 2026.
The AIMS still helps with many legal obligations, because it delivers the processes the law presupposes: roles, risk assessment, documentation, training, and an AI inventory as the basis for classification. Our comparison EU AI Act vs. ISO 42001: What Is Mandatory, What Is Good Practice? shows where the standard helps and where it does not.
The law's deadlines apply regardless of certification. The transparency obligations under Article 50 have applied since August 2, 2026, and our article on AI labeling under Article 50 covers the details.
Under Regulation (EU) 2026/1744, the high-risk obligations apply from December 2, 2027 for Annex III systems and from August 2, 2028 for Annex I systems. Our article on high-risk AI under the EU AI Act explains which systems are covered.
For the full picture of roles, risk classes, deadlines, and enforcement, read our guide to the EU AI Act.
ISO 42001 and ISO 27001: What you can reuse
If you run an ISMS under ISO 27001, you can reuse almost the entire management system framework: Clauses 4 to 10, the risk method, document control, internal audit, and management review. The new content mainly concerns how AI affects people and how AI systems move through their life cycle. Our guide to ISO 27001 certification sums up the basics of your ISMS.
| Element | Already in your ISO 27001 ISMS | New for the ISO 42001 AIMS |
|---|---|---|
| Context and scope (Clause 4) | Method, interested parties, scope document | Role as AI provider, AI producer, or AI user; AI systems in scope; AI inventory |
| Policy (5.2) | Information security policy | A dedicated AI policy (A.2) |
| Roles (5.3) | CISO, risk owners | Owners of AI systems, a channel for reporting concerns (A.3) |
| Risk assessment (6.1.2) | Method, risk register | AI risk sources from Annex C, risks to people outside the company |
| Impact assessment (6.1.4, 8.4) | Not covered; can build on the GDPR data protection impact assessment | Impact assessment per AI system (A.5) |
| Statement of Applicability (6.1.3) | 93 controls from ISO 27001 Annex A | A separate statement for the 38 controls in ISO 42001 Annex A |
| Data | Classification, deletion, access control | Provenance, quality, and preparation of training and test data (A.7) |
| Suppliers | Supplier relationships and cloud services | AI vendors, allocation of responsibilities along the chain (A.10) |
| Development | Secure development | Requirements, verification and validation, monitoring in operation (A.6) |
| Training (7.2, 7.3) | Security awareness training | AI-specific competence and training |
| Internal audit, management review (9.2, 9.3) | Process in place | Extend to the AIMS; one joint program is possible |
A second misconception concerns security. An ISO 42001 certificate shows that you govern AI, but not that an AI system can withstand attacks such as data poisoning or prompt injection. Since October 14, 2026, the guidance standard ISO/IEC 27090 has covered that ground: it describes nine types of compromise and offers no certification of its own.
Our overview of the AI security standard ISO 27090 explains how to add these threats to your risk register. For the most common one, we show separately how to assess prompt injection in an ISO 27001 ISMS.
If you also provide US customers with a SOC 2 report, you will meet AI there too. The AICPA's October 2026 draft of the Trust Services Criteria adds points of focus on AI governance, model monitoring, and training data. An AIMS supplies much of the evidence, as our article on the 2027 SOC 2 Trust Services Criteria shows.
The path to certification in five phases
ISO 42001 certification follows five phases: scope and AI inventory, risk and impact assessment, controls and Statement of Applicability, internal audit with management review, and finally the two-stage certification audit. With an existing ISMS, the first three phases take from under one to two months; without one, from three to four months. The internal audit and certification audit add about one month, or one to two months without an ISMS.
- Scope and AI inventory (owner: the AIMS lead, with IT and business teams). You determine your role under Clause 4.1 and record every AI system in scope: your own models, third-party models built into your product, and AI features in tools you already use. Unapproved tools are the blind spot here, and we explain separately how to uncover shadow IT in your company.
- AI risk assessment and impact assessment (the AIMS lead, with product and data protection). You assess risks against the risk sources in Annex C and, for each system, estimate how it could affect individuals, groups, and society. ISO/IEC 42005:2025 gives guidance on this, and ISO/IEC 23894:2023 on risk management.
- Controls, policies, and Statement of Applicability (the AIMS lead, with sign-off by management). You put the AI policy, roles, approval process, supplier assessment, and training in place, and you document in the Statement of Applicability which of the 38 controls apply.
- Internal audit and management review (internal or external auditors, management). The internal audit checks the AIMS against the standard, and the management review decides on objectives and resources (Clauses 9.2 and 9.3).
- Certification audit, Stage 1 and Stage 2 (certification body). In Stage 1, the certification body checks whether the AIMS is documented and ready for audit; in Stage 2, whether it works in daily operation.
Our article on ISMS certification under ISO 27001 walks through a Stage 1 and Stage 2 audit in detail. ISO 42001 follows the same logic.
Book your certification body early. Since July 2025, ISO/IEC 42006 has set the requirements for bodies that audit AI management systems. Several certification bodies active in Germany currently hold ISO 42001 accreditation through accreditation bodies abroad, for example TÜV NORD through UKAS and RvA, and DQS through ANAB. Many bodies offer combined audits with ISO 27001, provided they are accredited for both standards.
How the audit went at Kertos
Kertos built its AIMS in 2025 alongside its existing ISMS. The internal audit in early September 2025 focused on the AI inventory: every AI tool in use across the company had to be recorded, because unrecorded tools count as shadow AI in the audit. The AIMS covers the AI tools used internally as well as Kaia, the AI assistant in the Kertos platform.
The Stage 1 audit took place on November 4, 2025. Afterwards, the work that needed the most push was not the documentation but getting every employee to acknowledge the policies and complete their training. About a week later, Kertos passed Stage 2 with a single finding, and the certificate arrived at the end of 2025. The first surveillance audit is due in November 2026.
Two lessons follow. First, your scope includes the AI you only use, not just the AI you build. Second, an AIMS fails less often on paper than in the team: plan policy acknowledgments and training as separate work packages before Stage 1. The article Kertos achieves ISO 42001 certification describes how Kertos puts AI governance into practice internally.
Example: a SaaS company with 120 employees
The following figures are illustrative. A SaaS company with 120 employees has held ISO 27001 certification since 2024. It has built a summarization feature based on a hosted language model into its product and uses a support assistant internally. An enterprise customer asks for proof of AI governance at contract renewal.
The team sets the scope to cover the product and all AI tools used internally. For the product feature, the company is a provider; for the internal tools, it is a user. The inventory finds 14 AI systems, five of them AI features in existing tools that nobody had recorded as AI so far.
The CISO owns the project and spends about half of their working time building the AIMS. The head of product contributes to the impact assessment for the summarization feature, and the data protection officer aligns it with the existing data protection impact assessment. From kickoff in June to Stage 2 in October, the project takes just under five months, because the inventory and the product's technical documentation take longer than planned. The company combines Stage 2 with its ISO 27001 surveillance audit.
How Kertos supports you with ISO 42001
With Kertos, you build the AIMS on the same platform as your ISMS. The platform links the ISO 27001 and ISO 42001 controls, so you don't maintain shared requirements from Clauses 4 to 10 or existing evidence twice. You record AI systems in the AI inventory, assess their risks in the AI risk assessment, and link them to policies and controls.
You create policies in the platform's policy management, assign training to employees, and track completion as evidence. Kertos's certified experts support you from scoping through the impact assessment to the certification audit. Kertos holds ISO 42001 certification itself, so we know the requirements from our own audit. The page on ISO 42001 in the Kertos platform gives an overview of the features.
If you want to find out how much of ISO 42001 your existing ISMS already covers, book a demo and bring your AI use cases.
Frequently asked questions
What is ISO 42001?
ISO/IEC 42001 is the international standard for AI management systems (AIMS). It sets out how a company governs the development, provision, and use of AI through policies, roles, risk assessment, and impact assessment. Published in December 2023, it is the only standard against which an AI management system can be certified.
Is ISO 42001 mandatory?
No. ISO/IEC 42001 is voluntary, and the EU AI Act does not make it mandatory either, because it is not a harmonized standard. It only becomes binding if a customer requires it contractually or a tender makes it a prerequisite.
How long does ISO 42001 certification take?
With an existing ISO 27001 ISMS, certification is realistic in as little as 1 to 3 months; without one, in as little as 4 to 6 months. These figures assume full support within the company and are lower bounds. Common delays are unrecorded AI tools, missing training records, and the availability of the certification body.
What is the difference between ISO 42001 and ISO 27001?
ISO 27001 protects information through an information security management system with 93 controls in Annex A. ISO 42001 governs the use of AI through an AI management system with 38 controls and also considers how AI affects people and society. Both standards share the same structure in Clauses 4 to 10, and a certification body can audit them together.
Who can certify against ISO 42001?
Certification bodies accredited for ISO/IEC 42001 can issue certificates. ISO/IEC 17021-1 and, specifically for AI management systems, ISO/IEC 42006:2025 set the requirements for these bodies. Before you engage one, check which accreditation body granted the accreditation and whether it covers your scope.
