ISO/IEC 42001 explained: the AI management system standard for software teams

Clauses 4 to 10, control topics in our own grouping, and the line between management-system work and evidence a running system can record.

12 min read
In short: ISO/IEC 42001:2023 is the certifiable international standard for an AI management system, published by ISO and IEC in December 2023. Clauses 4 to 10 define the management system and are your organization's work. Annex A lists controls, and running systems can record evidence for some control topics, such as resources, the life cycle, use, and third parties. Arcjet maps its records to ISO/IEC 42001 and doesn't cover the clauses.

What is ISO/IEC 42001?

ISO/IEC 42001:2023 is the international standard for an AI management system: the policies, objectives, risk process, audits, and improvement cycle an organization uses to govern the AI it develops, provides or uses. The published abstract says the standard specifies requirements and gives guidance for establishing, implementing, maintaining and continually improving such a system, and that it applies to any organization that provides or uses products or services that use AI, whatever its size. Organizations can be certified against it. The requirements sit in clauses 4 to 10, and Annex A lists controls.

ISO and the International Electrotechnical Commission (IEC) published the standard in December 2023. The joint ISO/IEC committee on AI, JTC 1/SC 42, wrote it. The text is copyrighted, and you can buy it from iso.org or your national standards body. This guide paraphrases and never quotes it. For anything more detailed than this guide, such as the exact controls, use your licensed copy.

An AI management system is not software. It is the set of documented decisions, processes, and records that show an organization manages AI on purpose. This guide explains what the standard asks for at a high level and which parts a software team can evidence from systems it runs. To work through the control topics one by one, use the ISO 42001 Annex A checklist. For how the standard fits into a wider program, see what is AI governance.

Who gets certified against ISO/IEC 42001, and why?

Organizations that build, sell or use AI get certified when their customers, regulators or partners want independent assurance that AI is governed. Certification is voluntary. A certification body audits the management system and issues the certificate.

The most common reasons to certify are the following:

  • Customer due diligence. Enterprise buyers send AI questionnaires, and a certificate answers many of them at once.
  • An existing ISO program. If you already hold ISO/IEC 27001, the auditors, document control and internal audit process are in place.
  • Regulatory preparation. A management system gives you the risk process, records, and reviews that laws such as the EU AI Act expect. A certificate isn't a legal safe harbor, and it doesn't make you compliant with any law.

The audit follows the same pattern as other ISO management system certifications. Ask your certification body for its audit stages and how often it returns during the certificate's term. In each audit, the auditor checks that your processes exist and that records show they ran. That is where software-generated evidence helps.

How does ISO/IEC 42001 relate to ISO/IEC 27001?

ISO/IEC 42001 and ISO/IEC 27001 are both ISO management system standards, and their clauses 4 to 10 cover the same management topics. Teams that already run an information security management system (ISMS) can extend it: the same internal audit, management review, document control and corrective action processes can serve both.

The two differ in subject. ISO/IEC 27001 covers information security. ISO/IEC 42001 is meant to help an organization develop, provide or use AI systems responsibly. Each standard has its own Annex A controls.

In practice, many AI risks are also security risks. Prompt injection, a leaked secret in a prompt, or an agent that calls a destructive tool show up in both programs. One runtime record can serve as evidence for both, as long as you map it in each.

What do clauses 4 to 10 of ISO/IEC 42001 require?

Clauses 4 to 10 require you to define, run, check, and improve an AI management system, and they hold the requirements a certification audit checks. The following table summarizes the general purpose of each clause in our own words and lists the evidence an auditor usually expects. Read the AI-specific requirements in each clause in your copy of the standard.

ClauseWhat it is about (our summary)Typical evidence
4. Context of the organization

Work out what affects your AI management system, who has a stake in it, and its scope

A scope statement and a list of interested parties
5. Leadership

Top management commits to the system, sets an AI policy and assigns roles

An approved AI policy and named roles
6. PlanningAssess and treat AI risks, and set AI objectivesA risk assessment, a risk treatment plan and AI objectives
7. Support

Provide resources, competence, awareness, communication and controlled documents

Training records and a document register
8. OperationRun the processes you planned, including risk treatmentRecords that the planned processes ran
9. Performance evaluation

Monitor and measure, run internal audits and hold management reviews

Monitoring results, audit reports and review minutes
10. ImprovementFix nonconformities and improve the system continuallyCorrective action records

These clauses are management work. No product performs them for you. What software can do is supply records for clause 8 (the processes ran) and clause 9 (monitoring results).

Which control topics does an ISO/IEC 42001 program cover?

Annex A lists controls. Decide which ones apply from your risk assessment, and record why you exclude any. The standard sets the numbering and titles of its controls, so check them in your copy.

For a working list, the following table groups the control topics an AI management system program usually has to address. The grouping and the summaries are ours, not the standard's, so map each topic to the Annex A controls in your copy.

Topic (our grouping)What it covers (our summary)
AI policyA documented AI policy, aligned with other policies and reviewed
Roles and responsibilities

Who is accountable for each AI system, and how concerns are raised

Resources

The data, tools, compute and people each AI system depends on, documented

Impact assessmentHow each AI system affects people, assessed and recorded
Life cycle

Requirements, design, verification, deployment, operation, monitoring and retirement, managed as a life cycle

DataData quality, provenance and preparation
Information for othersDocumentation and information for users and others affected
UseResponsible use within each system's intended purpose
Third parties

Suppliers, customers and partners, and how responsibilities are shared with them

Resources, the life cycle, use and third parties are where running systems produce records. AI policy, impact assessment and information for others are documents and judgments that people produce. Roles are both: people assign them, and a system can record the assignment and keep it current.

What can software supply as ISO/IEC 42001 evidence?

Software can supply records that a process ran and that a control operates: who owns each AI system, which providers and tools it uses, which decisions controls made, and when systems were reviewed or retired. It can't supply policy, judgment, impact assessments, competence or disclosures to people outside the organization.

An AI system is an application or agent together with the model providers and tools it uses. The following table shows, for each control topic where runtime records help, what Arcjet records and what stays with you. The evidence page also maps these records to ISO/IEC 42001.

TopicWhat Arcjet recordsWhat stays with you
Roles and responsibilities

Owner per AI system, actor and provider; ownership lapses when the owner leaves the team

Concern reporting, and the competence of each owner
Resources

Providers, Model Context Protocol (MCP) servers and tools in use per AI system, suggested from observation and confirmed by people

Compute, data sets and staffing
Impact assessmentRisk tier recorded per AI systemThe impact assessment itself
Life cycle

A test before a rule goes live, dry run, a second approver to publish a policy, runtime records, and retired actors with keys revoked

Design, verification and release of the model or application
Data

Declared versus observed data classes, from sensitive-information findings

Data quality, provenance and labeling
Use

Each system's purpose linked to the Guard policies that enforce it, and purpose drift notices

Training users on intended use
Third parties

Provider decisions, enforcement on managed devices, and violations

Supplier assessments and contracts

Records like these answer an auditor's most frequent question, "show me that it happened", with data from the control rather than a screenshot. They don't answer "show me that you decided this was the right control". That is your risk treatment and your record of which controls you apply and why.

What Arcjet does not cover

Arcjet doesn't cover clauses 4 to 10. It isn't a management system, and it doesn't write your AI policy, run your internal audit or hold your management review. It supplies evidence for some ISO/IEC 42001 requirements, and the evidence page marks every other requirement as not covered.

The following topics get no evidence from Arcjet:

  • AI policy. Guard policies are runtime rules, not policy documents. Keep your AI policy in your document control system or governance, risk, and compliance (GRC) tool.
  • Impact assessment. Arcjet records a risk tier on each AI system. The assessment of impacts on people is yours.
  • Information for others. Documentation and disclosures for users and affected people are yours.

Within the topics Arcjet supports, the limits in the preceding table apply. The inventory also covers only what Arcjet observes: managed macOS and Windows devices with Discovery enrolled, actors that report through the SDK, and coding agents with hooks installed. The evidence page states inventory coverage as the share of enrolled devices that are healthy, so an auditor can see how complete it is.

Where Arcjet fits

An ISO/IEC 42001 audit asks for many of the same things other governance frameworks do: a list of the AI systems you run, a named owner and purpose for each, decisions about which providers are allowed, and evidence that those decisions hold. Most teams answer with a register that someone updates before the audit.

Arcjet builds that register from observation. Discovery finds the AI providers that people use on managed devices and enforces provider decisions there. The SDK and coding-agent hooks report the agents you build and run. Guard decides tool calls, prompts, and model use, records each decision, and requires a rule to pass a test before it goes live. Owner, purpose, risk tier and review date sit on each AI system, and every change to them is written to the WorkOS audit log.

The evidence page maps those records to ISO/IEC 42001, as well as to the NIST AI RMF and the EU AI Act. It lists every requirement and marks each as supported or not covered. Export it as JSON or CSV, or push it to Vanta or Drata, and attach it to your record of applied controls. For the steps, see how to export AI governance evidence.

To check each control topic in detail, continue with the ISO 42001 Annex A checklist. To compare the standard with the other two frameworks, see NIST AI RMF vs ISO 42001 vs EU AI Act.

Frequently asked questions

Is ISO/IEC 42001 certification mandatory?

No. Certification is voluntary. Organizations pursue it when customers, partners, or regulators want independent assurance that AI is governed. A certification body audits the AI management system and issues the certificate.

Can you get ISO 42001 certified without ISO 27001?

Yes. ISO/IEC 42001 stands alone. Holding ISO/IEC 27001 makes it easier, because both are ISO management system standards whose clauses 4 to 10 cover the same management topics, so internal audit, management review, document control, and corrective action processes can serve both.

Does ISO 42001 certification satisfy the EU AI Act?

No. A certificate shows you run an AI management system. It isn't a legal safe harbor under the EU AI Act. The management system does produce much of the risk process, records, and review that the Act expects, so the work overlaps.

Can a software tool make you ISO 42001 certified?

No. Certification covers an organization's management system: its policy, roles, risk process, audits, and reviews. Software can supply records that processes ran and controls operate, which auditors ask for, but the management system and the certificate are the organization's.

Where can I read the ISO/IEC 42001 text?

The standard is copyrighted. Buy it from iso.org or your national standards body. Articles such as this one paraphrase the clauses and control topics and don't reproduce the text.

AI runtime security in your code

Protect your AI agent workflows with Arcjet

Get allow, deny, and redact on agent actions before the side effect.