ISO/IEC 42001:2023 specifies requirements for an AI management system (AIMS): the policies, processes and controls an organisation uses to develop, provide or use AI responsibly. Published in December 2023, it is the first AI governance standard that organisations can be certified against, and certifications are now common among AI vendors who need to reassure enterprise customers.

Who it is for

Any organisation that develops, provides or uses AI systems, of any size. A bank deploying third-party AI tools and a start-up building models can both apply it. The scope is defined by your role and the AI systems involved.

Familiar structure

If you know ISO/IEC 27001, the structure will feel familiar. Both follow the common ISO management-system format (the "harmonised structure"):

  • Clause 4, Context: understand internal and external issues, interested parties, and your role (developer, provider or user) for each AI system. Define the scope.
  • Clause 5, Leadership: top management commitment, an AI policy, roles and responsibilities.
  • Clause 6, Planning: AI risk assessment, risk treatment, AI system impact assessment, and objectives.
  • Clause 7, Support: resources, competence, awareness, communication and documentation.
  • Clause 8, Operation: carry out the plans, assessments and treatments, and control changes.
  • Clause 9, Performance evaluation: monitoring, internal audit and management review.
  • Clause 10, Improvement: nonconformity, corrective action and continual improvement.

Because the structure is shared, many organisations run 42001 as an extension of an existing ISO 27001 ISMS, reusing document control, internal audit and management review.

What is distinctive

AI system impact assessment

Beyond risks to the organisation, 42001 asks you to assess the potential consequences of AI systems for individuals, groups and society: fairness, safety, privacy, transparency and other impacts. This is closer to a DPIA than to a classic information-security risk assessment. The companion standard ISO/IEC 42005 gives detailed guidance on doing it.

Annex A controls

Annex A lists reference controls for responsible AI. Annex B gives implementation guidance. The controls are grouped under objectives covering:

  • AI policies;
  • internal organisation (roles, reporting concerns);
  • resources for AI systems (data, tools, computing, people);
  • assessing impacts of AI systems;
  • the AI system life cycle (requirements, design, verification, deployment, operation, monitoring, documentation);
  • data for AI systems (quality, provenance, preparation);
  • information for interested parties (documentation, transparency, incident communication);
  • use of AI systems (responsible use, intended use, human oversight);
  • third-party and customer relationships (suppliers, allocation of responsibilities).

As with ISO 27001, you select applicable controls based on risk and record the decisions in a Statement of Applicability.

How it relates to other frameworks

  • EU AI Act: 42001 is not the same as the Act's legal requirements, but an AIMS supplies much of what high-risk obligations demand: risk management, data governance, documentation, logging, human oversight and post-market monitoring. Harmonised European standards are being developed for the Act specifically. See the EU AI Act after the Digital Omnibus.
  • NIST AI RMF: a voluntary framework with four functions (Govern, Map, Measure, Manage) that maps well onto 42001 and is useful for detailed practices.
  • ISO/IEC 23894: guidance on AI risk management, which complements clause 6.
  • ISO/IEC 27701: for privacy, alongside 42001 when AI processes personal data.

Getting started without a big programme

  1. Inventory AI use. Every AI system you build, buy or use, with an owner. Include generative AI tools staff use. See shadow AI.
  2. Write a short AI policy approved by leadership: principles, roles, approval routes and prohibited uses.
  3. Set up intake and triage. New AI use cases go through a simple form and a risk tiering. High-risk ones get a full risk assessment and impact assessment.
  4. Reuse what you have. Extend existing risk, supplier, change and incident processes to cover AI rather than building parallel ones.
  5. Pick a few measures, such as use cases assessed, incidents, model evaluations completed and training completion, and review them in management review.
  6. Decide on certification later. Build the system to be useful first. Certification makes sense when customers or regulators ask for it.

Learn more

AI governance frameworks, including ISO/IEC 42001 and the NIST AI RMF, are covered in our free AI Security & Governance course. For the security-management counterpart, see ISO/IEC 27001:2022 explained.

Written and reviewed by the DigiTrust Asia editorial team. How we research, review and correct articles.

General information, not legal advice. Laws and official guidance change; check the current text before acting.

Build AI governance skills

AI Security & Governance is a free course with a final exam and certificate.

Create a free account