Reliable Context Article
ISO/IEC 42001 Explained: What an AI Management System Is
What ISO/IEC 42001 requires of an AI management system: clause structure, impact assessment, documented information, certification, and audit evidence.
Summary
- 01ISO/IEC 42001 governs how an organization manages AI risks and opportunities. A certificate describes its management system; it does not establish that a particular model is accurate, fair or safe.
- 02The harmonized structure supports an integrated management system across standards. AI-specific content includes risk assessment, impact assessment, data governance and system lifecycle controls.
- 03The statement of applicability documents necessary controls and justifies inclusion or exclusion. Organizations may not need every Annex A control and may exceed its list.
- 04Policies describe what should happen; records show what happened, when and on whose authority. Auditable evidence must be retrievable, dated and attributable.
- 05Conformity can exist without certification. Certification is voluntary and carried out by independent bodies; reviewing a supplier claim also means checking accreditation and the certified scope.
Inside this article
- 01What ISO/IEC 42001 is
- 02What a management system standard is
- 03The structure: clauses 4 to 10
- 04The AI-specific content
- 05Documented information and records
- 06Relationship to other ISO/IEC standards
- 07Certification versus conformity
- 08Worked example: what an auditor might ask a data platform for
- 09Common misreadings
What ISO/IEC 42001 is
ISO/IEC 42001:2023, Information technology — Artificial intelligence — Management system, was published in December 2023 by the joint ISO and IEC subcommittee on AI, ISO/IEC JTC 1/SC 42 [1]. It "specifies the requirements and provides guidance for establishing, implementing, maintaining and continually improving an AI (artificial intelligence) management system within the context of an organization" [2]. It applies to any organization "that provides or uses products or services that utilize AI systems", whatever its size or type (same source).
The full text is paywalled. This article uses only public material: ISO's product and explainer pages and the free preview on the ISO Online Browsing Platform (OBP), which shows the introduction, scope, definitions, table of contents and bibliography. For clauses 4 to 10 it reports titles, not requirement text.
What a management system standard is
A management system is a "set of interrelated or interacting elements of an organization to establish policies and objectives, as well as processes to achieve those objectives" [2]. A policy is the intentions and direction of an organization as formally expressed by its top management. An objective is a result to be achieved (3.5, 3.6, same source).
A management system standard (MSS) therefore governs how an organization decides, assigns, records and corrects. It does not test a product. ISO says of 42001: "Rather than looking at the details of specific AI applications, it provides a practical way of managing AI-related risks and opportunities across an organization", and it is implemented "using the Plan‐Do‐Check‐Act methodology" [1].
For a procurement or risk team, this is the key point: a 42001 certificate describes a management system. It does not say a particular model is accurate, fair or safe.
The harmonized structure
ISO management system standards share a harmonized structure (HS). The ISO/IEC Directives describe it as "identical clause numbers in the same sequence, clause titles, text, common terms, core definitions" [3]. This lets one organization run a single "integrated" system against several standards [4]. The 42001 introduction states that it "applies the harmonized structure" [2].
The structure: clauses 4 to 10
Clauses 1 to 3 cover scope, normative references and terms; the requirements sit in clauses 4 to 10. The titles below come from the public table of contents [2]. The text under them is in the paid document.
| Clause | Title | Subclauses (titles only) |
|---|---|---|
| 4 | Context of the organization | 4.1 Understanding the organization and its context; 4.2 Understanding the needs and expectations of interested parties; 4.3 Determining the scope of the AI management system; 4.4 AI management system |
| 5 | Leadership | 5.1 Leadership and commitment; 5.2 AI policy; 5.3 Roles, responsibilities and authorities |
| 6 | Planning | 6.1 Actions to address risks and opportunities; 6.2 AI objectives and planning to achieve them; 6.3 Planning of changes |
| 7 | Support | 7.1 Resources; 7.2 Competence; 7.3 Awareness; 7.4 Communication; 7.5 Documented information |
| 8 | Operation | 8.1 Operational planning and control; 8.2 AI risk assessment; 8.3 AI risk treatment; 8.4 AI system impact assessment |
| 9 | Performance evaluation | 9.1 Monitoring, measurement, analysis and evaluation; 9.2 Internal audit; 9.3 Management review |
| 10 | Improvement | 10.1 Continual improvement; 10.2 Nonconformity and corrective action |
Because the standard applies the harmonized structure, much of this skeleton is shared with other management system standards. AI appears explicitly in the titles of 4.3, 5.2, 6.2 and 8.2 to 8.4. ISO's explainer adds "data governance and system lifecycle controls" to its summary of the requirements [5].
A management system is only as checkable as its records; the practical test for a data platform under an AIMS is whether a reader can trace a stated fact to its dated source and see who recorded it.
The AI-specific content
Why AI gets its own management system
The introduction gives three reasons AI "potentially raises specific considerations": AI can make automatic decisions "in a non-transparent and non-explainable way"; machine learning changes how systems are "developed, justified and deployed"; and systems that learn continuously "change their behaviour during use" [2]. Organizations "are expected to focus their application of requirements on features that are unique to AI", and they "can elect to apply these requirements using a risk-based approach" (same source).
The processes it names include managing trustworthiness concerns "such as security, safety, fairness, transparency, data quality and quality of AI systems throughout their life cycle", and managing "suppliers, partners and third parties that provide or develop AI systems for the organization" (same source).
Impact assessment
42001 defines an AI system impact assessment as a "formal, documented process by which the impacts on individuals, groups of individuals, or both, and societies are identified, evaluated and addressed" [2]. It has its own subclause, 8.4, separate from AI risk assessment in 8.2, and it points outward at people and society rather than at the organization. ISO/IEC 42005:2025 gives guidance on carrying out these assessments "throughout the AI system lifecycle" [6].
Controls, Annex A and the statement of applicability
42001 has four annexes: Annex A, "Reference control objectives and controls"; Annex B, "Implementation guidance for AI controls"; Annex C, "Potential AI-related organizational objectives and risk sources"; and Annex D, "Use of the AI management system across domains or sectors" [2]. The Annex A control table is not in the public preview, so its controls are not listed here. The Annex B subclause titles show its guidance topics:
- B.2 Policies related to AI
- B.3 Internal organization
- B.4 Resources for AI systems
- B.5 Assessing impacts of AI systems
- B.6 AI system life cycle
- B.7 Data for AI systems
- B.8 Information for interested parties
- B.9 Use of AI systems
- B.10 Third-party and customer relationships
A control is a "measure that maintains and/or modifies risk" [2]. A statement of applicability (SoA) is "documentation of all necessary controls ... and justification for inclusion or exclusion of controls" (3.26). Its notes add that organizations "may not require all controls listed in Annex A or may even exceed the list", and that "all identified risks and the risk management measures (controls) established to address them shall be reflected in the statement of applicability" (3.26, Notes 1 and 2).
Data
42001 defines data quality as the "characteristic of data that the data meet the organization's data requirements for a specific context", taken from ISO/IEC 5259-1 [2]. The definition is relative: data are good when they meet requirements the organization has stated for a specific use. The introduction says 42001 "avoids specific guidance on management processes" and leaves organizations to draw on other frameworks and standards for risk, life cycle and data quality management [2].
Documented information and records
Documented information is "information required to be controlled and maintained by an organization and the medium on which it is contained". It "can be in any format and media and from any source". It can refer to the management system itself, to documentation created to operate, and to "evidence of results achieved (records)" [2]. Clause 7.5 carries that title.
The distinction decides what an audit can check. A policy says what should happen; a record shows what did happen, when, and on whose authority. The introduction states: "An organization conforming with the requirements in this document can generate evidence of its responsibility and accountability regarding its role with respect to AI systems" (same source). Such evidence depends on records that can be retrieved, dated and attributed: in data terms, on an audit trail and on provenance for the values a decision relied on.
Relationship to other ISO/IEC standards
| Standard | What ISO says it does | Relation to 42001 |
|---|---|---|
| ISO/IEC 22989:2022 | Establishes AI terminology and describes AI concepts [7] | The only normative reference in 42001. Its terms apply alongside those in clause 3 [2] |
| ISO/IEC 23894:2023 | Guidance on managing risk specifically related to AI [8] | Guidance, not certifiable requirements. Listed in the 42001 bibliography |
| ISO/IEC 42005:2025 | Guidance for AI system impact assessments [6] | Supports the impact assessment in 8.4 |
| ISO/IEC 27001 | Requirements for an information security management system [9] | A sibling MSS on the same harmonized structure. ISO sells the two together |
| ISO/IEC 5259-1:2024 | Overview, terminology and examples for data quality in analytics and ML [10] | Source of the 42001 data-quality definition |
| ISO/IEC 42006:2025 | Additional requirements for bodies that audit and certify AI management systems [11] | Governs the certifiers, not the certified organization |
ISO/IEC 27001 covers information security, the "preservation of confidentiality, integrity and availability of information" [2], while 42001 covers AI. Sharing a structure, they can run as one integrated system; Annex D.2 of 42001 is titled "Integration of AI management system with other management system standards". 23894 and 42005 are guidance: certification is against 42001, not them.
The 42001 bibliography also lists the NIST AI Risk Management Framework 1.0 [2]. NIST released it on 26 January 2023 for voluntary use [12]; mapping the two is outside this article's scope.
Certification versus conformity
Conformity is "fulfilment of a requirement". A nonconformity is its non-fulfilment. An audit is a "systematic and independent process for obtaining evidence and evaluating it objectively to determine the extent to which the audit criteria are fulfilled", and it can be internal (first party) or external (second or third party) [2]. An organization can conform without being certified, and 42001 has its own internal audit subclause (9.2) either way.
Certification is a separate, voluntary step. ISO states that "ISO does not certify organizations" and that certification "is carried out by independent certification bodies, which may be accredited by national accreditation bodies" [5]. Accreditation is formal recognition that a certification body "operates according to international standards"; ISO points to the IAF CertSearch database to verify accredited certificates [13]. Two standards govern the certifiers:
- ISO/IEC 17021-1:2015 sets requirements for the "competence, consistency and impartiality" of management system certification bodies [14].
- ISO/IEC 42006:2025 "builds on ISO/IEC 17021-1" with AI-specific requirements for bodies certifying against 42001 [11].
Reviewing a supplier's claim, ask: which body issued the certificate, is it accredited for this scheme, and what is the certified scope? Scope matters because "organization" in 42001 refers only to the part of a larger entity "that is within the scope of the AI management system" [2]. A certificate may cover one business unit.
The EU AI Act
In the European Union, Regulation (EU) 2024/1689 (the AI Act) requires providers of high-risk AI systems to "put a quality management system in place". Article 17(1) lists aspects including "systems and procedures for data management", the risk management system, post-market monitoring, and "systems and procedures for record-keeping of all relevant documentation and information" [15]. Under Article 40(1), a presumption of conformity attaches to harmonised standards "the references of which have been published in the Official Journal of the European Union" (same source). As of 28 October 2026, the first harmonised standard referenced in the Official Journal under the AI Act is EN 18286:2026, Artificial intelligence - Quality management system for EU AI Act regulatory purposes, published through Commission Implementing Decision (EU) 2026/2274 and announced on 19 October 2026; it gives a presumption of conformity with the Article 17 requirements it covers, and the Commission describes it as designed to be layered on top of standards such as ISO/IEC 42001 rather than replace them [16]. ISO/IEC 42001 is not a harmonised standard under the AI Act, so 42001 certification should not be read as a presumption of conformity.
On timing, Regulation (EU) 2026/1744 (the Digital Omnibus on AI) entered into force on 27 July 2026. It sets the application of the high-risk requirements in Chapter III, Sections 1 to 3, to 2 December 2027 for systems classified under Article 6(2) and Annex III, and to 2 August 2028 for those under Article 6(1) and Annex I [17]. This describes the legal text and is not legal advice.
- Conformity means fulfilling a requirement.
- An organization can conform without being certified, and internal audit remains part of the standard.
- Certification is a separate, voluntary step.
- Independent certification bodies carry out certification and may be accredited by national accreditation bodies.
When reviewing a supplier claim, check the issuing body, its accreditation for the scheme and the certified scope. A certificate may cover one business unit.
Worked example: what an auditor might ask a data platform for
Fictional example. Halden Therapeutics AG uses a machine-learning model to prioritize incoming adverse-event reports for human review. It sets the scope of its AIMS (4.3) to cover that system. Its AI risk assessment (8.2) identifies a risk that reports with missing fields are ranked too low. The treatment (8.3) is a control that sends any report below a completeness threshold to a reviewer, and the control appears in the SoA with its justification. The impact assessment (8.4) records the possible effect on patients if a serious report is delayed.
Six months later, an internal auditor (9.2) samples one delayed report. The table shows records a data layer could supply; it is an illustration mapped to clause titles, not the standard's requirements.
| Auditor's question | Relevant clause title | Record a data layer can supply |
|---|---|---|
| Was this system inside the AIMS scope on that date? | 4.3 Determining the scope | Dated scope document with version history |
| Which control applied, and was it in force? | 8.3 AI risk treatment; SoA (3.26) | Control identifier, effective-from date, the SoA version that listed it |
| What input did the model see? | 7.5 Documented information | The report as received, with source, receipt time and every later correction |
| Who changed the completeness threshold, and when? | 6.3 Planning of changes | Append-only change log with author, timestamp and approval reference |
| Who could see or edit the record? | 5.3 Roles, responsibilities and authorities | Access grants by role, consistent with need-to-know access |
| Was the nonconformity found and fixed? | 10.2 Nonconformity and corrective action | Finding, root cause, corrective action and closure date, all linked |
Each row asks: can the record be reproduced as it stood on that date, with source and author? If the platform overwrote the input or threshold, the audit can confirm only the current state.
- 01Set the scope
Halden Therapeutics AG includes the system that prioritizes adverse-event reports for human review within its AIMS scope.
- 02Assess the AI risk
The risk assessment identifies that reports with missing fields may be ranked too low.
- 03Apply and document the control
Reports below a completeness threshold go to a reviewer. The control and its justification appear in the statement of applicability.
- 04Assess the impact on patients
The impact assessment records the possible effect on patients if a serious report is delayed.
Common misreadings
| Misreading | What the public sources say |
|---|---|
| "ISO certified us." | ISO does not certify. Independent certification bodies do [13] |
| "42001 certifies our model." | It is a management system standard and does not look at "the details of specific AI applications" [1] |
| "Every Annex A control is mandatory." | Organizations may not need all Annex A controls and may add their own. Inclusion and exclusion are justified in the SoA (3.26) |
| "It only applies to AI developers." | It applies to organizations that provide or use AI-based products or services (clause 1) |
| "42001 certification equals AI Act compliance." | The AI Act's presumption attaches to harmonised standards published in the Official Journal (Art. 40(1)). The first such standard is EN 18286:2026 (QMS, Article 17); 42001 is not one |
A management system is only as checkable as its records; the practical test for a data platform under an AIMS is whether a reader can trace a stated fact to its dated source and see who recorded it.
External Sources (17)
Published by
Reliable Context Research
Reliable Context is a system of record for organizations that fuse private and public information and must defend every value they act on: one canonical record per real thing, ambiguity shown rather than hidden, and provenance behind every value.
Editorial note
Researched and written by Reliable Context Research with AI-assisted research tools; sources are cited inline and checked before publication. Report an error: https://reliablecontext.com/corrections.