Annex A of ISO/IEC 42001:2023 on iso.org is the normative list of reference control objectives and controls for an AI management system, printed on pages 17 to 20 of the standard (official preview, table of contents). Published summaries group the controls under nine objective areas, A.2 to A.10, and commonly count 38 controls.
You do not have to apply every control. The standard defines the Statement of Applicability as the record of “all necessary controls” and the justification for including or excluding each one (definition 3.26). Your risk treatment decides which controls you need. Annex A is the list you check that decision against.
| Objective area | What it governs | Record it typically produces | Clause it supports | Controls in the crosswalk numbering |
| A.2 Policies related to AI | The AI policy, how it fits other policies, and its review | Approved AI policy with a review date | 5.2 AI policy | 3 |
| A.3 Internal organization | AI roles and a route for raising concerns | Roles and responsibilities record; concern reporting procedure and log | 5.3 Roles, responsibilities and authorities | 2 |
| A.4 Resources for AI systems | Data, tools, computing and people each AI system depends on | Resource documentation per AI system | 7.1 Resources | 5 |
| A.5 Assessing impacts of AI systems | Effects of AI systems on individuals, groups and society | Impact assessment procedure and completed, dated assessments | 6.1.4 and 8.4 AI system impact assessment | 4 |
| A.6 AI system life cycle | Responsible design, build, testing, deployment, operation and logging | Requirements, design records, verification and validation results, deployment approval, monitoring records, event logs | 8.1 Operational planning and control | 9 |
| A.7 Data for AI systems | Data acquisition, quality, provenance and preparation | Data records per system: sources, quality checks, provenance, preparation steps | 8.1, with 6.1.2 AI risk assessment | 5 |
| A.8 Information for interested parties | User information, external reporting and incident communication | User documentation, reporting channel, incident notices | 7.4 Communication | 4 |
| A.9 Use of AI systems | Responsible use and the intended use of each system | Responsible use procedure, use objectives, intended use statement | 8.1, with 4.1 intended purpose | 3 |
| A.10 Third-party and customer relationships | Who is responsible for what across suppliers and customers | Responsibility allocation, supplier assessments, customer terms | 4.2 interested parties and 8.1 | 3 |
Sources: objective area titles from the nine control categories in the IAPP’s June 2024 summary; A.2 to A.10 numbering and control counts from the draft-era (FDIS) crosswalk contributed by Microsoft and listed on NIST’s AI Resource Center; clause titles from ISO/IEC 42001:2023 (official preview). The record and clause columns are an AI Training Skool mapping, not wording from the standard.
What is ISO 42001 Annex A, and what is it not?
Annex A is a reference set of controls that an organization selects from, not a checklist it must complete. The standard’s own table of contents marks Annex A as normative, which means it is part of the requirements. Applying each individual control is still a decision your risk treatment makes.
The clauses work differently. Clauses 4 to 10 are mandatory for any organization that claims conformity, from scope (clause 4) to corrective action (clause 10). Annex A controls are the “how” you choose for the risks those clauses make you find. For building the clause side, follow the order of work in an ISO 42001 implementation.
Two notes to definition 3.26 set the rules for the control set. You may not need every Annex A control. You may also add controls of your own that Annex A does not list. All identified risks, and the controls that address them, “shall be reflected” in the Statement of Applicability.
The nine objective areas carry plain titles. The IAPP’s summary of the ISO/IEC 42001 controls (June 2024) names nine control categories: policies related to AI, internal organization, resources for AI systems, assessing impacts of AI systems, AI system life cycle, data for AI systems, information for interested parties, use of AI systems, and third-party and customer relationships.
How many controls are in ISO 42001 Annex A?
Published summaries commonly count 38 controls across nine objectives. ISO does not state the count on any free iso.org page, so treat 38 as the market’s reading of the standard, not an ISO-published figure.
The closest public source that numbers the controls is the NIST AI RMF to ISO/IEC 42001 crosswalk in NIST’s AI Resource Center crosswalk listing. The listing credits it to Microsoft, and NIST notes that listing a crosswalk does not imply endorsement. The document maps the AI RMF to the FDIS (final draft) text of ISO/IEC 42001. It cites 38 distinct numbered implementation items, B.2.2 to B.10.4, grouped under the same nine areas. Annex B gives implementation guidance for the Annex A controls under matching numbers, so the crosswalk numbering lines up with the 38 in published summaries.
| Objective area | Items in the crosswalk | Spread |
| A.2 | 3 | Policy, alignment, review |
| A.3 | 2 | Roles, concerns |
| A.4 | 5 | Documentation plus four resource types |
| A.5 | 4 | Process, documentation, individuals, society |
| A.6 | 9 | Two subgroups: development objectives and processes (6.1), then the life cycle steps (6.2) |
| A.7 | 5 | Development data, acquisition, quality, provenance, preparation |
| A.8 | 4 | Users, external reporting, incidents, interested parties |
| A.9 | 3 | Processes, objectives, intended use |
| A.10 | 3 | Responsibilities, suppliers, customers |
| Total | 38 | Nine objective areas |
Source: Microsoft-contributed crosswalk listed on NIST’s AI Resource Center (FDIS text, file dated May 2023), counted by AI Training Skool, October 2026. Control titles in the published standard may differ slightly from the FDIS.
If you are filling in a spreadsheet, build one row per control from your own licensed copy of the standard. If you do not hold one yet, see buying ISO/IEC 42001 from ISO or ANSI. Lists copied from blogs drift: some page-one results in October 2026 claimed 39.
Which Annex A controls set up governance (A.2 to A.4)?
A.2, A.3 and A.4 set up the governance base: a policy, named roles and a documented view of what each AI system runs on. They are the controls an auditor usually reads first, because every later control points back to them. The policy controls in this area build on the clause 5.2 AI policy, and a section-by-section AI policy skeleton shows one way to write it.
A.2 Policies related to AI
A.2 asks for an AI policy, alignment with your other policies, and review of the AI policy. The crosswalk titles the three items AI policy, alignment with other organizational policies, and review of the AI policy. The evidence is an approved policy document, a note of which existing policies it touches (security, privacy, quality) and a dated review record.
A.3 Internal organization
A.3 covers AI roles and responsibilities, and the reporting of concerns. The first produces a record of who is accountable for each AI system and each AIMS process. The second produces a channel where staff can raise a concern about an AI system, plus a log showing how concerns were handled.
A.4 Resources for AI systems
A.4 asks you to document the resources each AI system depends on. The crosswalk lists five items: resource documentation, data resources, tooling resources, system and computing resources, and human resources. In practice this is a resource section in each AI system register entry, kept current as models, vendors and teams change.
Impact assessment, life cycle and data controls (A.5 to A.7)
A.5, A.6 and A.7 are where the AI-specific work sits. Together they hold 18 of the 38 items in the crosswalk numbering, and they produce most of the records an AI system leaves behind.
A.5 Assessing impacts of AI systems
A.5 asks for an AI system impact assessment process, documentation of the assessments, and assessment of impacts on individuals or groups and on society. The standard defines the AI system impact assessment as a formal, documented process for identifying, evaluating and addressing those impacts (definition 3.24). That is wider than a privacy assessment. The records are the procedure and a dated assessment per system.
A.6 AI system life cycle
A.6 is the largest objective area, with nine items in two groups. The first group covers objectives and processes for responsible design and development. The second follows the life cycle step by step:
- Requirements and specification
- Documentation of design and development
- Verification and validation
- Deployment
- Operation and monitoring
- Technical documentation
- Recording of event logs
Each step leaves a record, from a requirements document to retained logs. If you buy rather than build AI, some steps fall to your supplier, and A.10 decides who holds which record.
A.7 Data for AI systems
A.7 covers the data used to develop and improve AI systems: acquisition, quality, provenance and preparation. For each system you should be able to show where its data came from, what quality checks ran, and how it was cleaned or labeled.
How A.8, A.9 and A.10 handle transparency, use and third parties
A.8, A.9 and A.10 face outward: what you tell people about an AI system, how it may be used, and who carries which responsibility across the supply chain.
A.8 Information for interested parties
A.8 lists system documentation and information for users, external reporting, communication of incidents, and information for interested parties. Expect user-facing documentation, a channel for outside parties to report problems, and a record of incidents you communicated.
A.9 Use of AI systems
A.9 covers processes and objectives for responsible use, and the intended use of each AI system. The evidence is a use procedure, measurable use objectives and an intended use statement per system. That statement is what lets you spot a system being used for a purpose nobody assessed.
A.10 Third-party and customer relationships
A.10 asks you to allocate responsibilities and to manage suppliers and customers. If you use AI you did not build, this area carries much of your evidence: supplier assessments, contract clauses and a responsibility split per system.
How to use Annex A with risk treatment and the Statement of Applicability
Use Annex A after you have assessed risk, not before. Pick controls to treat the risks you found, check the full Annex A list so nothing relevant is missed, and then record every decision in the Statement of Applicability.
- Assess AI risks (clause 6.1.2). Score each risk with the method you documented, such as an AI risk assessment method that survives an audit.
- Choose treatments (clause 6.1.3, AI risk treatment). Decide which controls reduce each risk. Some come from Annex A, and some may be your own.
- Check against Annex A. Walk all nine objective areas. A control you skip needs a reason tied to your risks or your role, not to effort.
- Write the Statement of Applicability. One row per control: included or excluded, the justification, and where the evidence lives. Before approval, check the rows against common Statement of Applicability nonconformities.
- Repeat in operation (clauses 8.2 and 8.3). The standard asks you to perform risk assessment and treatment in operation too, so the SoA changes when systems change.
- Test it (clause 9.2, internal audit). Sample SoA rows and check that the evidence exists.
Example only (not from a real organization): a US software company that uses a vendor’s large language model in a support chatbot might mark A.6.2.8, event logs, as included, because it must reconstruct what the bot told a customer. It might mark A.10.4, customers, as included too, because business clients ask for AI terms in contracts. If it never prepares training data itself, it might exclude A.7.6, data preparation, and write that reason in the justification column.
What Annex B, C and D add to Annex A
Annex B tells you how to implement the Annex A controls, Annex C lists objectives and risk sources to consider, and Annex D explains use across sectors. Only Annex B is normative alongside Annex A. Where the annexes sit in the whole standard is laid out in ISO 42001 structure, from clauses 4 to 10 to Annex D.
| Annex | Status | Pages | What it holds |
| A | Normative | 17 to 20 | Reference control objectives and controls |
| B | Normative | 21 to 45 | Implementation guidance for AI controls |
| C | Informative | 46 to 48 | Potential AI-related organizational objectives and risk sources |
| D | Informative | 49 to 50 | Use of the AI management system across domains or sectors |
Source: ISO/IEC 42001:2023 (official preview, table of contents).
Many summaries call Annex B informative. The official table of contents says normative. Standards Australia’s AS ISO/IEC 42001:2023 spotlight describes Annex B as implementation guidance for all Annex A controls. Annex B runs 25 pages against four for Annex A, so read the guidance before you write a control.
Annex C is a good prompt list for your risk assessment and AI objectives. Annex D matters if your AI management system has to sit next to sector rules. For more detailed methods on specific tasks, SC 42 publishes companion standards such as ISO/IEC 42005 for impact assessment and ISO/IEC 23894 for AI risk guidance. All appear in the ISO/IEC JTC 1/SC 42 published standards catalogue.
Is ISO 42001 Annex A the same as ISO 27001 Annex A?
No. The two standards come from different committees and manage different risks. ISO/IEC 27001 sets requirements for information security management systems and comes from ISO/IEC JTC 1/SC 27. ISO/IEC 42001 comes from JTC 1/SC 42, and its Annex A covers AI-specific matters such as impact on individuals and society, data provenance and intended use.
If you already hold ISO/IEC 27001, many A.4, A.7 and A.10 records can reuse existing evidence such as asset lists, supplier reviews and access logs. Keep a separate row for each 42001 control in your SoA even where the evidence is shared.
Who owns Annex A controls inside an organization?
The person running the AI management system owns the control set, and named system owners own the controls for each AI system. The US postings in the AI Training Skool AI Governance Pay Index (October 2026) include AI Governance Lead roles, and two postings that name ISO/IEC 42001 carry the titles Security GRC Lead and Senior AI Risk Analyst.
Annex A duties appear in a template AI governance job description in hiring language, such as maintaining an AI inventory and running risk and impact assessments. Will you be the person who selects controls and defends exclusions to an auditor? ISO/IEC 42001 Lead Implementer training covering clauses 4 to 10 and Annex A works through an AI register, impact assessment, risk register and Statement of Applicability end to end.
Frequently Asked Questions
How many controls does ISO 42001 have?
Published summaries commonly count 38 Annex A controls across nine objective areas, A.2 to A.10. ISO does not publish the count on a free iso.org page. A Microsoft-contributed crosswalk listed on NIST’s AI Resource Center, built against the final draft of ISO/IEC 42001, cites 38 numbered implementation items, B.2.2 to B.10.4, which matches that count.
Are all ISO 42001 Annex A controls mandatory?
No. Annex A is normative, but each control is selected through risk treatment. Definition 3.26 of ISO/IEC 42001:2023 says organizations may not need every Annex A control and may add their own. Every included or excluded control must appear in the Statement of Applicability with a justification, so an exclusion needs a reason tied to your risks.
What are the nine ISO 42001 control objectives?
The nine objective areas are A.2 policies related to AI, A.3 internal organization, A.4 resources for AI systems, A.5 assessing impacts of AI systems, A.6 AI system life cycle, A.7 data for AI systems, A.8 information for interested parties, A.9 use of AI systems, and A.10 third-party and customer relationships. The titles match the nine control categories in the IAPP’s June 2024 summary.
What is the difference between Annex A and Annex B in ISO 42001?
Annex A lists the reference control objectives and controls, on pages 17 to 20 of ISO/IEC 42001:2023. Annex B gives implementation guidance for those controls, on pages 21 to 45, under matching numbers. The official table of contents marks both as normative. Annex C and Annex D are informative.
What does ISO 42001 A.6 cover?
A.6 covers the AI system life cycle. It is the largest objective area, with nine items in the crosswalk numbering: objectives and processes for responsible development, then requirements, design documentation, verification and validation, deployment, operation and monitoring, technical documentation, and recording of event logs. Each step leaves a record an auditor can sample.
Is there an ISO 42001 controls list in Excel?
ISO does not publish an Excel controls list. Build your own spreadsheet from a licensed copy of ISO/IEC 42001:2023, with one row per Annex A control and columns for decision, justification, owner and evidence location. That spreadsheet becomes your Statement of Applicability. Free lists online vary, and some claim 39 controls.
