ISO/IEC 42001:2023 is an international standard for an AI management system: a managed way for an organization to set up, run and improve how it builds or uses AI. This glossary explains its main terms. It describes the framework only, and certification is a separate step done by an independent body.
What is the standard?
The IEC webstore lists ISO/IEC 42001:2023 as edition 1.0, published on 18 December 2023. It sets out requirements and guidance for establishing, implementing, maintaining and continually improving an AI management system within an organization, and applies to any organization, whatever its size, type or nature, that provides or uses products or services that use AI systems.
It follows the familiar management system pattern used by standards such as ISO/IEC 27001, so teams that run an ISMS will recognize the shape.
Core terms
AI management system (AIMS). The set of policies, processes, roles and records an organization uses to manage its AI work responsibly. It is a management system, not a tool.
AI system. The AI-based system that underpins a product or service the organization provides or uses. The standard covers both builders and users.
Interested parties. People or organizations with obligations to you or expectations of you: customers, regulators, employees, affected individuals.
Organization. The entity that sets up and runs the management system. Scope decides which parts of it the system covers.
Continual improvement. The standard expects you to review results and change the system over time, not set it once.
Control areas
The standard includes an annex of controls. We could not read the standard text, so the list below follows a third party summary from the Open Security Architecture project, whose own card says its titles are summaries and should be checked against the standard.
AI policy. A top-level statement of how the organization approaches AI, with roles and responsibilities.
AI risk assessment and treatment. Identifying risks from AI systems and deciding what to do about them.
AI impact assessment. Looking at how an AI system may affect individuals, groups or society. This is the area most unlike a pure security standard.
AI system lifecycle. Controls across design, development, acquisition, deployment, testing, change management, monitoring and logging.
Data for AI. Data quality, provenance and lineage, labelling, and integrity. Provenance means knowing where data came from and what happened to it.
Information for interested parties. Transparency, reporting and incident management.
Use of AI systems. Human oversight, user interaction and restrictions on system autonomy. Human oversight means a person can monitor and intervene.
Third-party relationships. Managing AI components and services from suppliers, including supply chain risk.
How does it relate to other frameworks?
It is not a replacement for security standards. An AI system still needs ordinary security controls. Teams often pair 42001 with ISO/IEC 27001 and map overlaps such as supplier management, logging and incident handling. It also differs from law: a management system standard is not a legal compliance guarantee, and regulations such as the EU AI Act have their own obligations.
What do first-time teams get wrong?
They write an AI policy and stop, so nothing changes in how models ship. They skip the AI system inventory, then cannot define scope. And they confuse a model card or a technical test report with an impact assessment, which looks at effects on people.
What to do
- List every AI system you build or use, including SaaS tools with AI features.
- Name an owner for each and decide scope.
- Draft a short AI policy with roles.
- Choose a risk method and assess your highest-impact system first.
- Add an impact assessment step to your release process for AI features.
- Record data provenance for training and retrieval data.
- Ask a certification body about timing only after the first internal review.
Quick check
Is ISO 42001 only for companies that build AI models?
No. Per the standard's scope, it applies to organizations that provide or use products or services that use AI systems.
Does using a security tool make me ISO 42001 certified?
No. Certification comes from an audit of your management system by an independent certification body. Tools can help gather evidence.
Sources
- IEC Webstore: ISO/IEC 42001:2023
- AFNOR: ISO/IEC 42001:2023
- Open Security Architecture: ISO/IEC 42001:2023 summary
Last reviewed: 18 October 2026.
