ISO/IEC 42001 is the first international management system standard for artificial intelligence. It gives organizations a structured way to govern how they develop, provide or use AI systems, and it can be audited and certified by independent bodies. This article explains what the standard asks for, how certification works and how to prepare.
What it is
ISO/IEC 42001 is titled "Information technology — Artificial intelligence — Management system". It was published jointly by the International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC) in December 2023, so the current edition is ISO/IEC 42001:2023.
It is a voluntary standard, not a law or regulation. Nothing in it is mandatory unless a customer, contract or regulator chooses to require it. It specifies requirements for an AI management system (AIMS): a set of policies, processes, roles and controls that an organization uses to manage AI-related risks and opportunities in a repeatable way.
Like ISO/IEC 27001 and other ISO management system standards, it follows the harmonized high-level structure (Clauses 4 to 10) and the Plan-Do-Check-Act improvement cycle. It also includes normative annexes with AI-specific controls and implementation guidance. Organizations can be certified against it by accredited certification bodies; ISO itself does not certify anyone.
The purpose is responsible AI: making sure AI systems are developed and used in ways that address concerns such as fairness, transparency, safety, accountability, privacy and security, while still supporting innovation.
Who it applies to
The standard is designed to be sector-agnostic and size-agnostic. It applies to any organization that provides or uses products or services that involve AI systems. That includes:
- AI developers and model providers, including those building machine learning features into products.
- Organizations that deploy or integrate third-party AI, such as SaaS vendors embedding models or enterprises using AI for hiring, underwriting or customer service.
- Public sector bodies and non-profits using AI in their operations.
There are no legal thresholds based on revenue or headcount. The trigger is practical: you have AI in scope and a reason to demonstrate governance, whether driven by customer due diligence, procurement requirements, regulatory alignment or internal risk management. You define the scope of your AIMS yourself, such as specific products, business units or the whole organization, and you document that scope.
Key requirements and control areas
Management system clauses
The main body of the standard requires, in summary:
- Context of the organization: understanding internal and external issues, interested parties and their requirements, and defining the AIMS scope and your role (for example, AI provider, producer or customer).
- Leadership: top management commitment, an AI policy, and assigned roles and responsibilities.
- Planning: AI risk assessment and risk treatment, AI system impact assessment, and measurable AI objectives.
- Support: resources, competence, awareness, communication and documented information.
- Operation: operational planning and control, running the risk assessments and treatments, and conducting impact assessments.
- Performance evaluation: monitoring and measurement, internal audit and management review.
- Improvement: handling nonconformities, corrective action and continual improvement.
Two features set it apart from a generic security management system: the AI risk assessment looks beyond information security to harms to individuals and society, and the AI system impact assessment evaluates the consequences of a system for people and groups affected by it.
Annex A controls
Annex A provides a reference set of controls organized under control objectives. As with ISO 27001, you compare your risk treatment against Annex A and produce a Statement of Applicability explaining which controls you include or exclude and why. The control areas cover:
- Policies related to AI, including alignment with other organizational policies.
- Internal organization: roles, responsibilities and reporting of concerns.
- Resources for AI systems: data, tooling, computing and human resources.
- Assessing impacts of AI systems on individuals, groups and society.
- AI system life cycle: design and development, verification and validation, deployment, operation and monitoring, and technical documentation.
- Data for AI systems: data management, quality, provenance and preparation.
- Information for interested parties: transparency, user information and incident communication.
- Use of AI systems: responsible use, intended use and human oversight.
- Third-party and customer relationships: supplier management, allocation of responsibilities and customer expectations.
Annex B gives implementation guidance for these controls. Further annexes address AI-related objectives and risk sources, and use of the AIMS across domains and with other management systems.
Assessment, audit and enforcement
Because ISO/IEC 42001 is a voluntary standard, there is no regulator enforcing it and no statutory penalties for not adopting it. Assurance comes from conformity assessment:
- Self-declaration: you can implement the standard and claim alignment, but this carries less weight with customers.
- Third-party certification: an accredited certification body audits your AIMS in stages. A Stage 1 audit reviews documentation and readiness; a Stage 2 audit tests implementation and effectiveness. If successful, a certificate is issued, followed by periodic surveillance audits and a recertification audit at the end of the certificate cycle.
- Internal audit and management review: required by the standard itself, and expected to be completed before the certification audit.
Certification bodies need to be accredited to certify against this standard, and the accreditation framework for AI management systems was still maturing after publication. Check that your chosen body holds the relevant accreditation. Auditors will look for documented evidence that processes operate as described, such as risk assessment records, impact assessments, review minutes, monitoring results and corrective actions.
A practical path to meeting it
- Define scope and role. Decide which AI systems, products and teams are in scope, and whether you act as provider, developer, deployer or a combination.
- Build an AI inventory. List models, datasets, vendors and use cases, including shadow AI and embedded third-party features, with an owner for each.
- Secure leadership commitment. Approve an AI policy, set objectives and assign accountable roles, such as an AI governance committee.
- Establish risk and impact methods. Define criteria for AI risk assessment and AI system impact assessment, then run them on in-scope systems.
- Select controls and write the Statement of Applicability. Compare your risks to Annex A, justify inclusions and exclusions, and plan treatment for gaps.
- Implement life cycle controls. Cover data quality and provenance, testing and validation, human oversight, documentation, monitoring and incident handling.
- Manage suppliers. Assess AI vendors and model providers, and document how responsibilities are split.
- Collect evidence and train staff. Make sure records are generated as part of normal work, not assembled retroactively.
- Run internal audit and management review. Close nonconformities before engaging the certification body.
- Undergo the certification audit and plan for ongoing surveillance, re-assessment when systems change and continual improvement.
How it maps to other frameworks
These are overlaps, not equivalences. Meeting one does not satisfy another.
- ISO/IEC 27001: the strongest overlap. Both use the same management system structure, so clauses on leadership, internal audit, management review and corrective action can often be shared. Annex A of 42001 is AI-specific and does not replace the 27001 information security controls. Many organizations integrate the two.
- NIST AI Risk Management Framework: a voluntary US framework organized around Govern, Map, Measure and Manage. The themes of risk identification, measurement and governance overlap with 42001, but the NIST framework is not certifiable and is less prescriptive about management system mechanics.
- EU AI Act: a binding regulation with obligations that depend on risk category and the organization's role. ISO/IEC 42001 can support parts of the required risk management, documentation and governance practices, but certification does not by itself demonstrate legal compliance. How standards relate to the Act's requirements was still developing at the time of writing.
- SOC 2: an attestation based on Trust Services Criteria. Security, confidentiality and processing integrity controls can share evidence with 42001, but SOC 2 has no AI-specific governance or impact assessment requirements.
- ISO/IEC 27701: privacy information management overlaps where AI processes personal data, particularly on data governance.
FAQs
Is ISO/IEC 42001 mandatory?
No. It is a voluntary international standard. It may become a practical requirement through customer contracts, procurement rules or as supporting evidence for regulatory obligations, but it is not a law.
Does ISO 42001 certification make us compliant with the EU AI Act?
No. The EU AI Act is a regulation with its own legal obligations. The standard can help you build governance, risk management and documentation processes that support compliance, but you still need to assess your specific obligations under the Act.
Do we need ISO 27001 first?
It is not a prerequisite. Organizations that already have an ISO 27001 management system usually find implementation easier because processes like internal audit and document control can be extended, but you can pursue 42001 independently.
What is the difference between the AI risk assessment and the AI system impact assessment?
The risk assessment addresses risks that could affect achieving the organization's AI objectives, including potential consequences to the organization, individuals and societies. The impact assessment focuses on the consequences of an AI system for individuals, groups and society. The standard requires both, and they inform each other.
How long does certification take?
It depends on your starting maturity, the number of AI systems in scope and existing management systems. Organizations with established ISO programs move faster; those without governance for AI should expect to build inventory, assessment and evidence processes first.
How Aristiun helps
Aristiun is the AI-powered compliance automation and threat modeling platform, powered by Ayurak AI. For ISO/IEC 42001 work, it helps keep the AIMS evidence-driven rather than a one-off documentation exercise. See the ISO 42001 framework page for more detail.
- Automated evidence collection: gather evidence automatically from connected systems such as AWS, Azure, GCP and GitHub, without manual screenshots or spreadsheets.
- Cross-framework mapping: map controls across 340+ frameworks so one evidence base can address ISO 42001, ISO 27001, the NIST AI RMF and the EU AI Act together.
- Gap analysis and monitoring: AI-powered validation identifies gaps early, and alerts flag controls that drift out of compliance before auditors find them.
- Threat modeling and vendor risk: generate threat models from code, IaC and architecture diagrams, and map vendor controls to your compliance requirements.