- ISO/IEC 42001:2023 is the international standard for an AI management system (AIMS). You can implement it on its own or have it certified by an accredited certification body
- It applies to organisations that develop, provide or use AI systems, and uses the same harmonised structure as ISO 27001 and ISO 9001
- It requires a documented process for AI system impact assessments; you set the criteria for when an assessment is needed and how deep it goes
- Annex A lists 38 AI-specific controls in nine groups (A.2 to A.10), selected through risk treatment and recorded in a Statement of Applicability
- ISO 42001 supports EU AI Act work but does not show legal compliance: duties such as Article 26 log retention and the Article 27 FRIA need separate checks
- As a planning estimate, organisations with a mature ISO 27001 system often allow 6 to 12 months; scope and the AI systems involved drive the real timeline
What ISO 42001 is
ISO/IEC 42001:2023, formally titled "Information technology, Artificial Intelligence, Management System", is the international standard for establishing, implementing, maintaining, and improving an Artificial Intelligence Management System (AIMS). It was published by ISO in December 2023 and became the first globally recognised certifiable standard for AI governance.
The standard is not a technical specification for building AI systems. It is a management system standard: it defines the governance processes, documentation, oversight mechanisms, and continual improvement cycles that an organisation needs in order to manage AI responsibly. It is concerned with how you make decisions about AI, not with the mathematics or engineering behind the AI itself.
ISO 42001 applies to any organisation that develops AI systems, provides AI-based products or services, or uses AI systems in its operations. That means software vendors building AI features, SaaS companies whose platforms incorporate AI, and businesses that deploy AI tools from third-party providers, all of them are within the scope of the standard. The obligations scale with the role: providers face more detailed implementation requirements, but deployers have real governance duties too.
Two forces are driving uptake. First, the EU AI Act requires documented processes for AI risk management, human oversight, and monitoring that ISO 42001 provides the framework to implement. Second, enterprise procurement, insurers, and investors are beginning to ask for evidence of structured AI governance, in the same way they asked for ISO 27001 a decade ago. The certification market for ISO 42001 is growing rapidly.
The standard's structure
ISO 42001 uses the Harmonised Structure (HS), the common framework shared by all modern ISO management system standards including ISO 27001, ISO 9001, and ISO 14001. This means that if your organisation already operates an ISO 27001 Information Security Management System, the governance architecture for ISO 42001 will be immediately familiar, and the two systems can be integrated rather than run in parallel.
The ten clauses of the standard follow the Plan-Do-Check-Act cycle:
The AI system impact assessment
The AI system impact assessment is the most distinctive element of ISO 42001. Clause 6.1.4 requires a process for assessing the potential consequences of AI systems for individuals, groups of individuals and society, and Clause 8.4 requires you to carry it out at planned intervals or when significant changes are proposed. The standard does not prescribe one template or one depth. You set documented criteria for when an assessment is needed and how detailed it must be, based on the system's intended use, the people affected and the risks identified.
In practice, an assessment looks at:
- Individuals who interact with the system or whose data it processes: accuracy, fairness, transparency and the possibility of harmful outcomes.
- Groups affected by the system's outputs at scale, including discriminatory or disproportionate effects.
- Society, where relevant: for example effects on access to services, public trust or the environment.
Risks to the organisation itself belong in the AI risk assessment (Clause 6.1.2), which the impact assessment feeds. A general writing assistant and a system that scores creditworthiness should not get the same depth of assessment. Record the result, use it to choose controls, and reassess when a change affects its validity: a new purpose, new data sources, a different model or a change in who uses the system.
ISO 42001 and the EU AI Act FRIA. An ISO 42001 impact assessment can provide useful input to the fundamental rights impact assessment (FRIA) that Article 27 of the EU AI Act requires of certain deployers of high-risk systems, such as public bodies, private entities providing public services and deployers that use AI for creditworthiness or for life and health insurance pricing. The two are not equivalent. Where Article 27 applies, check its required content and procedure separately, including notifying the market surveillance authority of the results.
Annex A controls
Annex A lists 38 AI-specific controls in nine groups. You select controls through AI risk treatment (Clause 6.1.3) and record the decisions in a Statement of Applicability, with a justification for each exclusion.
| Reference | Group | What it covers |
|---|---|---|
| A.2 | Policies related to AI | The AI policy, its alignment with other policies and its review |
| A.3 | Internal organisation | Roles, responsibilities and reporting of concerns |
| A.4 | Resources for AI systems | Data, tooling, system, computing and human resources |
| A.5 | Assessing impacts of AI systems | The impact assessment process and its documentation |
| A.6 | AI system life cycle | Objectives, design, verification and validation, deployment, operation, monitoring, technical documentation and event logs |
| A.7 | Data for AI systems | Data acquisition, quality, provenance and preparation |
| A.8 | Information for interested parties | Documentation for users, external reporting and communication of incidents |
| A.9 | Use of AI systems | Processes and objectives for responsible use, and use as intended |
| A.10 | Third-party and customer relationships | Allocating responsibilities with suppliers, partners and customers |
Annex B gives implementation guidance for each Annex A control. Annex C lists potential AI-related organisational objectives and risk sources, and Annex D covers use of the management system across domains and sectors.
ISO 42001 vs ISO 27001: what's different
| Dimension | ISO 27001 | ISO 42001 |
|---|---|---|
| Focus | Protecting information and information systems | Governing AI responsibly across its lifecycle |
| Risk scope | Confidentiality, integrity, availability of information | AI system impact on people, society, and the organisation; fairness, transparency, accountability |
| Key assessment | Information security risk assessment | AI system impact assessment + AI risk assessment |
| Annex A controls | 93 controls across 4 themes | 38 AI-specific controls in 9 groups (A.2 to A.10) |
| Scope definition | Information assets and systems in scope | AI systems in scope and the organisation's role for each |
| Certification cycle | 3-year cycle with surveillance audits and recertification | 3-year cycle with surveillance audits and recertification |
| Integration | Integrates with ISO 9001, ISO 14001, others | Designed to integrate with ISO 27001 and ISO 9001 |
For organisations already operating an ISO 27001 ISMS, the path to ISO 42001 is substantially shorter. The management system infrastructure, documented information, internal audit, management review, corrective action, is already in place. What needs to be added is AI-specific: the AI policy, impact assessment process, AI risk assessment, and the AI-specific controls from Annex A. Many organisations choose to integrate these into their existing ISMS rather than running a separate AIMS.
ISO 42001 and the EU AI Act
ISO 42001 and the EU AI Act deal with the same problem from different angles. The Act sets legal obligations for specific roles and risk levels. ISO 42001 gives you a management system that makes governance repeatable and auditable. A well-run AIMS contributes to several deployer obligations for high-risk AI, which apply to Annex III systems from 2 December 2027, but each obligation still needs its own check:
| EU AI Act requirement | ISO 42001 contribution | What you still need to check |
|---|---|---|
| Article 26: human oversight | Roles, competence and operational governance | Named people with the competence, authority and support to oversee each high-risk system |
| Article 26: monitoring | Monitoring, measurement and review (Clause 9) | Monitoring in line with the provider's instructions for use, and reporting of serious incidents |
| Article 26: logs | Documented information and record control | Keeping the automatically generated logs under your control for at least six months, or longer where other law requires |
| Article 26: informing workers | Communication and awareness | Informing workers' representatives and affected workers before a high-risk system is used at work |
| Article 27: FRIA | Impact assessment method | Whether Article 27 applies to you, its required content and notifying the authority |
ISO 42001 certification does not by itself give a legal presumption of conformity with the AI Act. Under Article 40, that presumption comes only from harmonised standards whose references are published in the Official Journal, and only for the requirements they cover. ISO 42001 is not on that track. The closest European standard is EN 18286, quality management systems for the purposes of the AI Act, approved by CEN-CENELEC in July 2026. As of October 2026 the Commission had not cited it, or any other AI Act standard, in the Official Journal, so check the current status before relying on it. How far an ISO 42001 system goes towards EN 18286 needs a requirement-by-requirement gap analysis.
2023
Getting started with ISO 42001
The path depends on where you start. Organisations with ISO 27001 already have the management system basics: documented information, internal audit, management review and corrective action. Organisations without a formal management system have further to go, but can build both standards on a shared base. ISO 42001 can also be implemented without seeking certification.
Step 1: Inventory your AI and define the AIMS scope
List the AI systems your organisation develops, provides or uses, with the purpose, the owner and your role for each (developer, provider or user). Keep the inventory and the formal scope apart: the inventory should cover AI use across the business, while a first certification scope can be a defined subset of activities. Systems you develop or provide need more detailed documentation and controls than third-party tools you only use.
Step 2: Set criteria and run AI system impact assessments
Define when an impact assessment is required and how deep it must go, then assess the systems that meet those criteria. Record who could be affected, how, and which safeguards apply. This step often exposes gaps: AI tools in use without review, data practices that need updating and oversight that is informal or missing.
Step 3: Run your AI risk assessment
Using the impact assessment results, assess AI risks to the organisation from failures, misuse or harmful outputs. Choose a treatment for each (accept, mitigate, transfer or avoid), document the decisions and link them to the Annex A controls you will implement.
Step 4: Set your AI policy and governance
Write an AI policy covering responsible use, human oversight, transparency towards affected people and data governance. Define clear responsibilities: who owns the AIMS, who reviews AI system performance and who runs the impact assessment process. AI governance can sit alongside an existing information security function, but it needs a named owner.
Step 5: Implement controls and prepare for certification
Implement the Annex A controls your risk treatment selected and document your Statement of Applicability. Run at least one internal audit and a management review before you engage a certification body. Initial certification usually has two stages: Stage 1 reviews your documentation and readiness, and Stage 2 tests whether the controls work in practice. Surveillance audits follow during the three-year cycle, with recertification before the certificate expires, under the rules of the certification body and its accreditation.
The minimum evidence to have ready
| Evidence | What it should show |
|---|---|
| AIMS scope | The activities and AI uses covered |
| AI policy | Leadership commitments and governance principles |
| AI system inventory | Systems, purposes, owners and your role |
| AI risk assessment and treatment | Criteria, results and treatment decisions |
| Impact assessment records | Affected parties, effects, safeguards and reviews |
| Statement of Applicability | Selected controls and justified exclusions |
| Supplier governance | Due diligence and responsibilities for third-party AI |
| Performance evaluation | Monitoring results, internal audit and management review |
| Corrective actions | How deficiencies are found and fixed |
Document names vary. What matters is that each item exists and stays current.
How long this takes depends on the scope, the number and criticality of your AI systems, your existing governance and the gaps you find. As a planning estimate, organisations with a mature ISO 27001 system often allow 6 to 12 months. The standard sets no timeline, and a mature ISMS does not by itself give you skills in model evaluation, data governance or impact assessment.
Our audits and compliance team supports organisations with ISO 42001 gap assessments, impact assessment design and implementation planning. Our AI advisory practice brings knowledge of AI system risks in financial services, healthcare and professional services.