Why it exists
Customers, boards and regulators now ask the same questions about AI: what is it used for, who is accountable, how do you know it's fair and accurate, and what happens when it goes wrong? ISO 42001 gives a structured, auditable answer. Published in December 2023, it is quickly becoming the reference point in enterprise procurement for AI products.
What an AI management system is
An AI management system is the set of policies, roles, processes and records an organisation uses to govern its AI. It isn’t a technical standard for building models, and it doesn’t tell you which algorithms to use. It requires you to know which AI systems you develop, provide or use, decide what responsible use means for each, assess the risks and the impact on people, put controls in place, and check and improve them over time.
The standard cares about your role. Drawing on the AI concepts and terminology in ISO/IEC 22989, it distinguishes AI providers, AI producers that design and build systems, AI customers and users, partners such as data providers and integrators, and the people affected by AI outputs. A company can hold several roles at once: provider of the AI feature in its product, and user of a third-party AI coding assistant. The controls differ for each.
Who should consider it
- SaaS and software companies with AI features in their product
- AI vendors selling to enterprise or government
- Businesses using AI in decisions about people: hiring, lending, insurance, healthcare
- Anyone receiving AI governance questions in security questionnaires
If AI is a minor internal productivity tool and nobody is asking, a lighter AI policy and register is often enough for now. Certification earns its place when AI is central to what you sell, affects decisions about people, or buyers ask for it by name.
What's in the standard
Clauses 4 to 10 mirror ISO 27001’s clauses: context, leadership, planning, support, operation, performance evaluation and improvement. The AI-specific parts are:
- AI risk assessment and AI system impact assessment, considering effects on individuals and society
- An AI policy and defined roles and responsibilities
- 38 Annex A controls across nine areas, from data quality and provenance to the AI system lifecycle and third-party relationships
- A Statement of Applicability, just as in ISO 27001 (see what an SoA contains)
The clauses, briefly
| Clause | What it adds for AI |
|---|---|
| 4 Context | Your role or roles in relation to AI, the intended purpose of each AI system, and the AIMS scope |
| 5 Leadership | An AI policy and top management accountability for AI governance |
| 6 Planning | AI risk assessment and treatment, the AI system impact assessment (6.1.4) and AI objectives |
| 7 Support | Resources, competence and awareness for people working on or with AI |
| 8 Operation | Running risk assessments, treatment and impact assessments at planned intervals and after significant change |
| 9 and 10 | Monitoring, internal audit, management review and corrective action, as in ISO 27001 |
The clause-by-clause comparison with ISO 27001 is in our ISO 42001 vs ISO 27001 guide.
The annexes
Annex A lists 38 controls in nine areas, numbered A.2 to A.10:
- A.2 Policies related to AI
- A.3 Internal organisation, including roles and reporting of concerns
- A.4 Resources for AI systems: data, tooling, computing and people
- A.5 Assessing impacts of AI systems on individuals, groups and society
- A.6 The AI system life cycle, from requirements and design to verification, deployment, operation and monitoring
- A.7 Data for AI systems: quality, provenance and preparation
- A.8 Information for interested parties, such as users and affected people
- A.9 Use of AI systems, including responsible use and intended purpose
- A.10 Third-party and customer relationships
Annex B gives implementation guidance for each control, so ISO 42001 carries its own equivalent of ISO/IEC 27002. Annex C lists potential AI-related organisational objectives and risk sources, a useful prompt when building your risk assessment. Annex D covers using the management system across domains and sectors.
The supporting standards
Three related standards are worth knowing. ISO/IEC 23894 gives guidance on AI risk management. ISO/IEC 42005:2025, published in May 2025, gives guidance on AI system impact assessments, the part most teams find hardest. ISO/IEC 42006:2025, published in July 2025, sets the extra requirements for certification bodies that audit ISO 42001, much as ISO/IEC 27006-1 does for ISO 27001. You certify only against ISO 42001; the others help you do it well.
How it fits with ISO 27001
| ISO 27001 | ISO 42001 | |
|---|---|---|
| Manages | Information security risk | AI risk and impact |
| Annex A | 93 controls | 38 controls |
| Shared | Management system clauses, internal audit, management review, document control, supplier management | |
Many organisations run one integrated management system covering both, with one internal audit programme and combined certification audits. The ISO 42001 vs ISO 27001 guide covers which to do first and how to run one integrated audit.
The regulatory backdrop
Australia
Australia has no AI-specific law that requires ISO 42001. In October 2025 the National AI Centre published the Guidance for AI Adoption, which updates and replaces the 2024 Voluntary AI Safety Standard and condenses its ten guardrails into six essential practices: decide who is accountable, understand impacts and plan accordingly, measure and manage risks, share essential information, test and monitor, and maintain human control. It is voluntary, and an ISO 42001 management system covers each practice with something an auditor has checked.
Existing law still applies to AI. From 10 December 2026, privacy policies must disclose when personal information is used in automated decisions that could significantly affect people; our guide to Australia’s new cyber and privacy laws explains the obligation. Consumer, anti-discrimination and sector rules apply to AI-driven decisions too. Government buyers increasingly ask suppliers how they govern AI, and an ISO 42001 certificate is a clean answer.
The EU AI Act
The EU AI Act applies to providers and deployers of certain AI systems, including some organisations outside the EU whose systems are used there. Its prohibitions and general-purpose AI model obligations already apply. Under amendments adopted in 2026, the main high-risk obligations now apply from 2 December 2027 for stand-alone high-risk systems and 2 August 2028 for AI embedded in regulated products. ISO 42001 supports much of the risk and quality management the Act requires, but certification does not by itself make you compliant with it.
Getting certified
The path matches ISO 27001: scope, gap assessment, implementation, internal audit, management review, then Stage 1 and Stage 2 audits by an accredited certification body. We implement ISO 42001 and run the internal audit. See our ISO 42001 page, or what to expect at Stage 1 and Stage 2.
In practice the work runs in this order:
- Inventory and scope. List the AI systems you develop, provide or use, your role for each, and which ones the AIMS covers.
- Policy and objectives. Agree what responsible AI means for your business and who is accountable.
- Risk and impact assessments. Assess each in-scope system for risks to the organisation and impacts on people, then choose controls and record them in the Statement of Applicability.
- Operate the controls. Data quality checks, verification and validation before release, monitoring in production, user information, supplier due diligence.
- Check and improve. Internal audit, management review and corrective action, then the certification audits.
Certification bodies need accreditation for ISO 42001 specifically. In Australia and New Zealand, JAS-ANZ runs an accreditation scheme for AI management systems, so check that a body’s accreditation scope names ISO/IEC 42001 before you engage it. As with ISO 27001, the certificate runs on a three-year cycle with surveillance audits in between; our guide to the ISO 27001 audit cycle explains how that works.
What auditors look for
- An AI inventory that matches reality, including AI features inside bought software
- Impact assessments that consider who could be affected and how, not a generic template
- Evidence that models were tested before release and are monitored after it
- Clear information for users, such as telling people when they’re dealing with AI
- Due diligence on AI suppliers and model providers
- Management review that actually discusses AI risks and performance
What drives the effort
Effort depends less on company size than on the number and nature of AI systems in scope, whether you build models or use them, how much of your training and input data you can trace, how many systems affect decisions about people, and whether you already have a working ISO 27001 ISMS to extend. An organisation with ISO 27001 and two AI features has far less to build than one starting from scratch with a portfolio of models. For how scope drives effort in a management system project, see what drives the cost of ISO 27001.
Common mistakes
- Treating ISO 42001 as an IT security project and leaving out product, data science and legal
- Scoping only the AI you build and forgetting the AI you use
- Writing an AI policy nobody follows, with no link to how models are actually released
- Copying the ISO 27001 risk method without adding criteria for impacts on people
- Assuming an ISO 27001 certification body can certify ISO 42001 without checking its accreditation
Next step
Start with an honest inventory of the AI you build and use, and the decisions it touches. That tells you whether you need certification now, a lighter governance framework, or ISO 27001 first. To check whether your AI use needs it, book a free 30-minute consultation.
