ISO/IEC 42001, explained

ISO/IEC 42001 is the international standard for an artificial intelligence (AI) management system. The International Organization for Standardization (ISO) catalog page gives the full title as Information technology - Artificial intelligence - Management system, the publication date as December 2023, the edition as the first, the length as 51 pages, and the responsible committee as ISO/IEC JTC 1/SC 42. ISO describes it as the world's first AI management system standard. The standard specifies requirements for establishing, implementing, maintaining, and continually improving an AI management system within an organization that provides or uses AI-based products or services. The text itself is sold, at 225 Swiss francs on ISO's site, so this post describes only what ISO's pages, the EU AI Act text, and a public crosswalk state.

What a management system standard is

ISO's management system standards page defines a management system as the way an organization manages the interrelated parts of its business to achieve its objectives, whether those objectives concern quality, efficiency, environmental performance, workplace safety, or something else. A management system standard (MSS) sets out requirements or guidance for those policies and processes. ISO says it publishes more than 80 of them, and that they are designed to apply across sectors, sizes of organization, and geographies.

The distinction matters for AI. A management system standard does not test a product. The standard asks whether the organization has policies, assigned roles, risk processes, documentation, audits, and management reviews in place and working. ISO's FAQ on the 42001 page puts it this way: rather than looking at the details of specific AI applications, the standard provides a practical way of managing AI-related risks and opportunities across an organization, using the Plan-Do-Check-Act method. The same FAQ defines an AI management system as a set of interrelated or interacting elements of an organization intended to establish policies and objectives, as well as processes to achieve those objectives, in relation to the responsible development, provision, or use of AI systems.

ISO's page also places 42001 among its other AI documents. ISO/IEC 22989 establishes terminology and concepts, ISO/IEC 23053 describes a framework for AI systems using machine learning, and ISO/IEC 23894 gives guidance on AI-related risk management for organizations. Those are guidance and vocabulary. ISO/IEC 42001, the FAQ says, "is a management system standard," the kind an organization can be audited against.

The Annex SL structure shared with ISO 27001

ISO's management system standards share a skeleton. The management system standards page explains that the Harmonized Structure (HS) means management standards are structured in the same way regardless of domain, so a user familiar with one feels at ease with another. Annex SL goes further: in addition to the same structure, there are some parts of a standard where identical text can be used, and the standards can contain many of the same terms and definitions. ISO says this is particularly useful for organizations that run a single, integrated management system meeting two or more standards at once, and points readers to the Joint Technical Coordination Group for management system standards for the current Annex SL text.

ISO/IEC 27001:2022, the information security management system standard, is the comparison most buyers know. ISO's page gives its third edition as published in October 2022, 19 pages, from ISO/IEC JTC 1/SC 27, with a 2024 amendment on climate action. ISO sells 42001 and 27001 together as an "AI and information security management package," which is a commercial signal of how ISO expects them to be used: side by side, under one integrated system.

The clause numbering shows the shared structure. The crosswalk hosted on NIST's AI Resource Center, discussed below, names 42001 clauses including 4.1 Understanding the organization and its context, 4.3 Determining the scope of the AI management system, 4.4 AI management system, 5.1 Leadership and commitment, 5.2 AI policy, 5.3 Roles, responsibilities and authorities, 6.1.2 AI risk assessment, 6.1.3 AI risk treatment, 6.2 AI objectives and planning to achieve them, 7.1 Resources, 7.2 Competence, 7.3 Awareness, 7.4 Communication, 7.5.3 Control of documented information, 8.2 AI risk assessment, 8.3 AI risk treatment, 8.4 AI system impact assessment, 9.1 Monitoring, measurement, analysis and evaluation, 9.2.2 Internal audit programme, and 9.3.2 and 9.3.3 on management review inputs and results. Anyone who has worked with 27001 will recognize clauses 4 through 9 in that list as context, leadership, planning, support, operation, and performance evaluation. The AI-specific additions sit inside that frame: an AI policy in clause 5, AI risk assessment and treatment in clauses 6 and 8, and an AI system impact assessment, which the crosswalk lists under clauses 6.1.4 and 8.4.

That last item has its own companion standard. ISO/IEC 42005:2025, published May 2025, 39 pages, provides guidance for organizations conducting AI system impact assessments, focused on how AI systems and their foreseeable applications may affect individuals, groups, or society. ISO sells it with 42001 as a "Responsible AI governance and impact standards package."

The controls, as public summaries describe them

ISO's catalog page does not list the annexes or controls, and the standard text is paid. Two public sources describe the content in outline.

ISO's own FAQ on the catalog page says the standard is designed to cover the various aspects of AI and the different applications an organization may be running, and provides an integrated approach to managing AI projects, from risk assessment to effective treatment of those risks. The page's list of benefits names a framework for managing risk and opportunities, demonstrating responsible use of AI, and traceability, transparency, and reliability. ISO's page for ISO/IEC 42006, the companion standard for certification bodies, says AI systems present unique challenges in areas like ethics, data quality, risk, and transparency, which is a fair summary of what an auditor is checking the management system for.

The second source is the NIST AI RMF to ISO/IEC FDIS 42001 crosswalk hosted on NIST's AI Resource Center. The file's metadata dates it to May 23, 2023, and the title says it maps to the FDIS, the final draft before publication, so numbering in the published standard may differ. For each AI RMF subcategory, the crosswalk lists the 42001 clauses and the implementation guidance items that correspond. The guidance items it names, grouped by their numbering, give the best public outline of what the control annexes cover:

Read against the controls literature, these are the familiar objects of a management system: a policy, named owners, inventories of resources, an impact assessment, life cycle documentation, data governance, disclosures, and supplier management, each applied to AI. What the published standard requires for each, and how an organization may exclude a control, is in the paid text and not on any public page, which is why vendor summaries that give exact control counts should be checked against the standard before they are repeated in a proposal.

Certification and who audits

ISO's certification page defines the two terms a buyer needs. Certification is the provision by an independent body of written assurance, a certificate, that the product, service, or system in question meets specific requirements. Accreditation is the formal recognition by an independent accreditation body that a certification body operates according to international standards. The page states that ISO does not perform certification or issue certificates, does not permit anyone to use the ISO logo in connection with certification, and that an organization therefore cannot be certified by ISO. Certification is performed by external certification bodies.

For ISO/IEC 42001, the chain has three links.

The organization seeking a certificate implements the management system. ISO's management system standards page notes that certification is not a requirement; an organization can implement the standard without being certified.

A certification body audits it. ISO/IEC 42006:2025, published July 2025, 31 pages, sets out the additional requirements for bodies that audit and certify AI management systems according to ISO/IEC 42001. ISO says it builds on ISO/IEC 17021-1, the general standard for management system certification bodies, and is meant to ensure that certification bodies operate with the competence and rigour needed to assess organizations developing, deploying, or offering AI systems. ISO's reasoning on the 42006 page is that auditors need specialised knowledge and clear rules for conducting assessments before they can certify that an organization manages AI's challenges responsibly.

An accreditation body assesses the certification body. ISO's certification page advises an organization choosing a certification body to evaluate several, to check that the body uses the relevant conformity assessment standard, and to check whether it is accredited, since accreditation provides independent confirmation of competence. The page adds that accreditation is not compulsory and that non-accreditation does not necessarily mean a body is not reputable, so a certificate from an unaccredited body is possible and should be weighed accordingly. ISO tells anyone verifying a certificate to use the International Accreditation Forum's CertSearch database, which aggregates data from accreditation bodies and certification bodies, or to contact the certification body or accreditation body directly. The page also notes that the Global Accreditation Cooperation has assumed the former roles of the International Accreditation Forum and the International Laboratory Accreditation Cooperation.

ISO 42001 vs the NIST AI RMF

The National Institute of Standards and Technology (NIST) AI Risk Management Framework, covered in the post on the NIST AI RMF, is a voluntary framework of outcomes organized into Govern, Map, Measure, and Manage. No one is certified to it. ISO/IEC 42001 is a requirements standard with a certification scheme. The two are built to be mapped, and NIST's AI Resource Center crosswalks page lists a crosswalk from the AI RMF to ISO/IEC 42001 submitted by Microsoft, with no date on the listing. NIST states that listing a crosswalk does not mean NIST endorses the mapped resource or that either resource fully covers the other, and that to be listed a crosswalk must map to a freely available resource, accurately reflect the AI RMF, and not advertise a product.

The crosswalk itself shows the grain of the mapping. Govern 1.1, on understanding legal and regulatory requirements, maps to clause 4.1 on context, clause 6.2 on AI objectives, and the guidance items on the AI policy and its review. Govern 1.3, on determining the needed level of risk management from risk tolerance, maps to clauses 6.1.1 through 6.1.3 on risk assessment and treatment. Govern 2.3, on executive responsibility for AI risk decisions, maps to clause 5.1 on leadership and commitment and the management review clauses in 9.3. Govern 1.6, on inventorying AI systems, maps to the resource documentation items. The pattern is that one AI RMF subcategory usually lands on several 42001 clauses and guidance items, and the crosswalk's title, which names the FDIS, is a reminder that the mapping predates the published text.

For a buyer, the practical difference is evidence. A vendor can claim alignment with the AI RMF and show a profile; a vendor can hold a 42001 certificate and show the certificate, its scope, and the accreditation of the body that issued it. Neither document says anything about a specific model's behavior.

ISO 42001 and the EU AI Act

The European Union's AI Act, Regulation (EU) 2024/1689, treats standards as a route to compliance, but only standards of a particular kind. Article 40, as the European Commission's AI Act Service Desk presents it, provides that high-risk AI systems and general-purpose AI models which conform to harmonised standards whose references have been published in the Official Journal of the European Union are presumed to conform with the corresponding requirements, to the extent those standards cover them. The article directs the Commission to issue standardisation requests to the European standardisation organisations covering all of the high-risk requirements, and says participants in standardisation should take into account existing international standards in the field of AI that are consistent with Union values and fundamental rights. The Service Desk notes that its text is the consolidated version as of July 27, 2026, including the amendments of the Digital Omnibus on AI.

ISO/IEC 42001 is an international standard, not a harmonised European standard, so a certificate to it does not by itself trigger Article 40's presumption. What it may do is cover much of what Article 17 requires of providers of high-risk systems: a quality management system, documented in policies, procedures, and instructions, with 13 listed elements that include a compliance strategy, design and development controls, testing and validation procedures, data management, the Article 9 risk management system, post-market monitoring, serious incident reporting, record-keeping, resource management, and an accountability framework. Article 17 says these aspects shall be proportionate to the size of the provider's organisation, while keeping the rigour needed for compliance. For providers that are financial institutions, paragraph 4 lets compliance with financial services governance rules satisfy most of the obligation and says harmonised standards under Article 40 shall be taken into account.

Timing matters for anyone planning around the Act. Article 113, in the consolidated text, says the regulation applies from August 2, 2026 in general, that Chapters I and II applied from February 2, 2025, except for some Article 5 prohibitions added by the Digital Omnibus, which apply from December 2, 2026, that the chapter on general-purpose AI models, the governance chapter, and the penalties chapter applied from August 2, 2025, and that the high-risk requirements in Chapter III, Sections 1 to 3, apply from December 2, 2027 for Annex III systems and from August 2, 2028 for Annex I systems. A 42001 program started now would be in place before the high-risk obligations bite.

What certification does and does not prove

A certificate proves that an accredited, or unaccredited, certification body audited an organization's AI management system against ISO/IEC 42001 on a date and found it conforming within a stated scope. The limit follows from what a management system standard is. ISO's definition covers the way an organization manages its policies and processes to achieve objectives, and ISO's FAQ for 42001 says the standard provides a way of managing AI-related risks across an organization rather than looking at the details of specific AI applications. What is verified is the organization's management system, including its processes for the development, provision, or use of AI.

A certificate therefore does not prove that a model is accurate, unbiased, or safe on a buyer's documents. The standard's clauses concern policy, roles, risk and impact assessment, and documentation; the audit checks that those exist and operate. The crosswalk names guidance items on verification and validation and on operation and monitoring; those titles describe processes for testing, not a particular test result. A certificate does not fix the model version either: the system is certified, and the organization may change what it builds within the system's scope.

A certificate does not prove regulatory compliance. ISO's certification page says certification is a useful tool to add credibility and that for some industries it is a legal or contractual requirement, but the AI Act's presumption of conformity attaches to harmonised standards in the Official Journal, and the NIST AI RMF has no certification at all.

And a certificate's value depends on who issued it. ISO's page says accreditation is not compulsory. ISO/IEC 42006 exists because, in ISO's words, auditors of AI management systems need specialised knowledge and clear rules. A buyer who sees a certificate should ask for the certification body's accreditation, the certificate's scope statement, and its dates, and should check the certificate in CertSearch.

What to check

• Does the certificate name ISO/IEC 42001:2023, the certification body, the accreditation body, the issue and expiry dates, and a scope that includes the AI system being bought?
• Is the certification body accredited for 42001 by a national accreditation body, and does the certificate verify in the IAF CertSearch database?
• Which clauses or controls did the organization exclude from its scope, and are the exclusions relevant to the deployment?
• Does the vendor's documentation of verification, validation, and operation and monitoring include results for the model version offered, or only the processes that would produce them?
• If the vendor also claims NIST AI RMF alignment, does it provide a crosswalk or profile, and which AI RMF subcategories does it mark as not covered?
• If the system will be sold in the EU as high-risk, how does the vendor plan to meet Article 17 and the harmonised standards that Article 40 presumes, since a 42001 certificate alone does not supply the presumption?

Azcuna's ELLMENT, the E-2D Large Language Model Entity, was proposed to NAVAIR PMA-231 under Navy SBIR Phase I topic DON26BZ01-NV010. The prototype, built at private expense on publicly available doctrine, answers with a page-level citation and refuses when the documents do not support an answer. The proposal has been submitted.

Questions people ask

What is ISO 42001?

ISO/IEC 42001:2023 is an international standard, titled Information technology - Artificial intelligence - Management system, that specifies requirements for establishing, implementing, maintaining, and continually improving an artificial intelligence management system within an organization. ISO's catalog page gives the publication date as December 2023, the first edition, 51 pages, and the committee as ISO/IEC JTC 1/SC 42. ISO calls it the world's first AI management system standard. The text is sold, not free.

What is an AI management system standard?

A management system, in ISO's words, is the way an organization manages the interrelated parts of its business to achieve its objectives. A management system standard sets requirements for that system rather than for a product. ISO's FAQ for 42001 defines an AI management system as a set of interrelated or interacting elements of an organization intended to establish policies, objectives, and processes for the responsible development, provision, or use of AI systems. Implementing the standard, the same FAQ says, uses the Plan-Do-Check-Act method.

How is ISO 42001 related to ISO 27001?

Both are ISO management system standards. ISO's management system standards page says these standards share a Harmonized Structure and, through Annex SL, can share identical text and many of the same terms and definitions, so an organization familiar with one can run an integrated system covering both. ISO/IEC 27001:2022 covers information security management and was published in October 2022; ISO sells the two together as a package. The crosswalk on NIST's site shows 42001 clauses such as 4.1, 5.1, 6.1.2, and 9.3 with the familiar numbering.

What are the ISO 42001 Annex A controls?

ISO's public pages do not list the controls, and the standard text is paid. ISO's FAQ says the standard provides an integrated approach to managing AI projects, from risk assessment to effective treatment of those risks, and ISO's page for the companion ISO/IEC 42006 names ethics, data quality, risk, and transparency as the areas assessed. The crosswalk hosted by NIST names guidance items on AI policy, roles, data and tooling resources, impact assessment, life cycle stages, data provenance and quality, information for interested parties, responsible use, and suppliers and customers.

What is ISO/IEC 42001 certification?

Certification is the provision by an independent body of written assurance that the product, service, or system in question meets specific requirements, as ISO's certification page defines it. For ISO/IEC 42001, the system is the organization's AI management system. ISO itself does not perform certification or issue certificates; external certification bodies do. ISO's page for ISO/IEC 42006, published July 2025, says those bodies work to ISO/IEC 17021-1 plus the additional requirements of 42006 when they audit and certify AI management systems.

Who audits and certifies ISO 42001?

External certification bodies audit organizations, and accreditation bodies assess the certification bodies. ISO/IEC 42006:2025, published July 2025, sets the additional requirements for bodies that audit and certify AI management systems, building on ISO/IEC 17021-1, so that auditors have the specialised knowledge ISO says AI systems demand. ISO's certification page defines accreditation as formal recognition by an accreditation body that a certification body operates to international standards, says accreditation is not compulsory, and points verifiers to the IAF CertSearch database.

Is ISO 42001 certification mandatory?

No. ISO's management system standards page states that certification to management system standards is not a requirement and that an organization can benefit from implementing a standard without being certified. A customer or contract can require a certificate. Under the EU AI Act, Article 40 gives a presumption of conformity only to harmonised standards whose references are published in the Official Journal, which is a separate status from an ISO publication.

How does ISO 42001 compare with the NIST AI RMF?

The NIST AI Risk Management Framework is a voluntary framework of outcomes organized as Govern, Map, Measure, and Manage, and no one certifies to it. ISO/IEC 42001 is a requirements standard an organization can be audited against. NIST's AI Resource Center hosts a crosswalk, submitted by Microsoft, that maps each AI RMF subcategory to clauses and guidance items of the standard, while stating that a listed crosswalk is not a NIST endorsement and that neither document fully covers the other.

Does ISO 42001 satisfy the EU AI Act?

Not by itself. Article 40 of the AI Act says high-risk AI systems and general-purpose AI models that conform to harmonised standards published in the Official Journal are presumed to conform with the corresponding requirements, and directs the Commission to issue standardisation requests to the European standardisation organisations. Article 17 requires providers of high-risk systems to have a quality management system with 13 listed elements. ISO/IEC 42001 is an international standard, and Article 40 says participants should take existing international standards into account.

What does an ISO 42001 certificate not prove?

A management system standard, in ISO's words, sets out requirements or guidance to help an organization manage its policies and processes; no product is tested. The certificate attests that the organization's AI management system met the standard at audit, which says nothing about a specific model's accuracy, bias, or safety on a buyer's task. ISO's certification page adds that accreditation of the certification body is not compulsory, so a buyer should check that the body is accredited and that the certificate's scope covers the AI system being bought.

Sources

1. 1ISO/IEC 42001:2023 Information technology - Artificial intelligence - Management system · International Organization for Standardization
2. 2Management system standards · International Organization for Standardization
3. 3Certification · International Organization for Standardization
4. 4ISO/IEC 27001:2022 Information security, cybersecurity and privacy protection - Information security management systems - Requirements · International Organization for Standardization
5. 5ISO/IEC 42006:2025 Requirements for bodies providing audit and certification of artificial intelligence management systems · International Organization for Standardization
6. 6ISO/IEC 42005:2025 AI system impact assessment · International Organization for Standardization
7. 7Crosswalks · NIST Trustworthy and Responsible AI Resource Center
8. 8NIST AI RMF to ISO/IEC FDIS 42001 AI Management system Crosswalk · NIST Trustworthy and Responsible AI Resource Center (submitted by Microsoft)
9. 9Article 40: Harmonised standards and standardisation deliverables · European Commission, AI Act Service Desk
10. 10Article 17: Quality management system · European Commission, AI Act Service Desk
11. 11Article 113: Entry into force and application · European Commission, AI Act Service Desk

Where this shows up in our work

• CALIBER, the market testbed for AI agents measures from an AI agent's trades whether it is gullible, herds, overpays, or can be steered.
• ELLMENT, the offline LLM for E-2D crews answers a crew's question from the aircraft's own manuals and cites the page, with no network.