Framework guide / ISO/IEC 42001:2023

ISO/IEC 42001: a practical implementation guide.

ISO/IEC 42001:2023 is the international standard for establishing, implementing, maintaining, and continually improving an Artificial Intelligence Management System (AIMS). It connects organizational AI policy, responsibilities, risk and impact assessment, operational controls, internal audit, and continual improvement into a management system that can be independently certified.

The implementation challenge is not producing more AI governance documents. It is demonstrating that the organization can identify what it governs, make accountable decisions, operate necessary controls, and respond when evidence shows those controls are insufficient.

Understand the standard
Explore the implementation toolkit

The implementation test

Can you demonstrate
how AI is actually governed?

A defined management boundaryIdentify the AI activities, responsibilities, and systems within scope.
Risk-based controls and evidenceConnect assessments, applicability decisions, controls, and operating records.
A functioning improvement processUse audit findings, management review, incidents, and changes to drive action.

An AIMS should connect organizational commitments to decisions and evidence that can be examined.

By InfoSecured
Source review: 17 October 2026
Sources & approach

01 / What ISO 42001 is

AI governance as an operating management system.

ISO/IEC 42001 specifies the requirements for an AI management system rather than a universal set of technical tests for individual AI models. It applies to organizations that develop, provide, or use AI-based products or services, regardless of organizational size or sector. [1]

An AIMS establishes how the organization governs relevant AI activities: who is responsible, what objectives apply, how risks and impacts are assessed, which controls are necessary, how performance is evaluated, and what happens when something must change.

Publisher & edition: ISO and IEC / ISO/IEC 42001:2023First edition, published December 2023. The published 2023 edition is the basis for this guide.
Standard type: AI management system requirementsOrganizational requirements that support implementation, internal assessment, and independent management-system certification.
Who it addresses: AI developers, providers, and usersResponsibilities and applicable controls depend on the organization’s role, its AI activities, its scope, and its external obligations.
What certification covers: A defined organizational management-system scopeCertification concerns the AIMS against applicable requirements. It is not blanket approval of every AI model, output, product, or deployment.

Organizations with existing information security, quality, privacy, or enterprise-risk processes may be able to reuse appropriate governance mechanisms. But an established process is useful evidence only when it actually addresses the AI activities within the AIMS scope.

Important distinction: ISO/IEC 42001 is a management system standard. Certification, regulatory compliance, and technical AI-system assurance are related but different conclusions.

02 / Clauses 4–10

Seven connected areas of management-system requirements.

The operational requirements in Clauses 4–10 cover the structure and operation of the AIMS. They are meant to work together: an approved policy without corresponding operational controls is incomplete, while an isolated technical assessment cannot substitute for leadership, scope, audit, and improvement.

04 / Context: Define the organization and AIMS boundary.Understand internal and external issues, relevant interested parties, the organization’s AI activities, and the scope of the management system. Establish how the AIMS will operate.
05 / Leadership: Establish policy, direction, and accountability.Secure top-management commitment, establish an AI policy, and assign responsibilities and authorities. Leadership accountability is not satisfied merely by naming a technical AI owner.
06 / Planning: Assess AI risk, impacts, and treatment needs.Define risk assessment and treatment arrangements, assess AI-system impacts, establish objectives, and plan changes. Select necessary controls and document their applicability.
07 / Support: Provide resources and controlled information.Address competence, awareness, communication, resources, and documented information. Records must remain usable and appropriately controlled throughout the management-system lifecycle.
08 / Operation: Put planned processes into practice.Operate and control the processes needed for the AIMS, perform risk and impact assessments as required, implement risk treatment, and respond to material changes.
09 / Performance evaluation: Determine whether the AIMS works.Monitor and evaluate relevant performance, conduct planned internal audits, and carry out management reviews informed by actual results and unresolved issues.
10 / Improvement: Correct problems and improve the system.Respond to nonconformities, determine appropriate corrective action, examine effectiveness, and improve the AIMS rather than merely recording that an action was assigned.

These requirements create an important distinction between designed governance and operating governance. A policy can describe the intended process. Evidence from actual assessment, approval, monitoring, audit, and corrective action shows whether the process has been used.

03 / Risk, impact & treatment

Connect assessments to a justified management decision.

ISO/IEC 42001 distinguishes AI risk assessment, AI risk treatment, and AI-system impact assessment. These activities inform each other but do not answer identical questions. The standard addresses them in Clauses 6.1.2, 6.1.3, and 6.1.4.

AI risk assessment / 6.1.2: What can affect the objectives, and how significant is it?Establish risk criteria and a repeatable assessment method. Consider the relevant AI activities, credible failure scenarios, uncertainty, consequences, and the conditions under which risk could materialize.
AI-system impact assessment / 6.1.4: What consequences could the system create for people and society?Examine intended use, foreseeable misuse, technical and societal context, affected individuals or groups, and relevant jurisdictions. Document results and consider them within AI risk assessment.
Risk treatment / 6.1.3: What response is justified by the assessment?Select treatment options and necessary controls, compare them against Annex A, document applicability decisions, and establish a treatment plan with appropriate management approval and residual-risk acceptance.
Reassessment / 8.2–8.4: When do the conclusions need to be reconsidered?Operate the relevant assessment and treatment processes, including reassessment at planned intervals or in response to significant changes under the applicable requirements.

Consider an AI system that summarizes customer complaints. A risk assessment may examine the possibility of misclassification and the organization’s inability to detect it. An impact assessment may examine the consequences for customers whose urgent complaints are overlooked. Risk treatment then determines which operational restrictions, evaluations, monitoring, escalation arrangements, or other controls are necessary.

These records should remain connected. A risk decision that ignores a material impact finding is difficult to defend, even if both documents were formally completed.

04 / Annex A controls

Annex A is a reference set, not a universal checklist.

Annex A contains 38 reference controls organized under nine subject areas, A.2–A.10. The standard requires an appropriate risk-treatment and control-selection process. It does not mean every Annex A control must be implemented in exactly the same way by every organization.

A.2 / AI policies: Set and maintain organizational direction.Align AI governance policy with relevant organizational commitments and ensure policies remain suitable.
A.3 / Internal organization: Clarify responsibility and reporting.Define relevant roles and make it possible for concerns about AI activities to reach appropriate decision-makers.
A.4 / AI-system resources: Understand the resources the system depends on.Consider data, technical tools, infrastructure, people, and other resources needed for development or operation.
A.5 / AI-system impacts: Assess consequences beyond technical performance.Establish processes and records for assessing relevant impacts on individuals, groups, and society.
A.6 / AI-system lifecycle: Govern development and operation.Address relevant lifecycle activities such as requirements, design, verification, deployment, monitoring, documentation, and event records.
A.7 / AI-system data: Address the data dependencies of AI.Consider relevant data sourcing, preparation, provenance, quality, and management activities.
A.8 / Information for interested parties: Provide appropriate system information.Address documentation, reporting, communication, and information needs for relevant users and other interested parties.
A.9 / Use of AI systems: Govern intended and actual use.Establish responsible-use processes and objectives and make the intended use of AI systems clear.
A.10 / Third-party relationships: Address responsibility across organizational boundaries.Consider supplier, customer, and other external relationships that affect AI governance or control operation.

The organization should identify the controls necessary for its assessed risks, compare them with the Annex A reference set, and consider whether additional controls are needed. Annex B supplies implementation guidance; Annexes C and D provide further informative material.

The Statement of Applicability

The Statement of Applicability, or SoA, records the necessary controls, the rationale for their selection, relevant exclusions, and implementation status. It is a central link between assessed risk and the organization’s actual control arrangements.

For example, an organization that only consumes an externally hosted AI service may not conduct model training itself. That can affect the rationale for particular development-related controls. It does not remove the organization’s own responsibilities for intended use, supplier dependencies, impacts, oversight, or applicable external requirements.

Control-selection rule: A justified exclusion is not the same as a missing control. Conversely, marking a control applicable does not establish that it is adequately implemented or effective.

05 / Evidence & records

Make requirements traceable to implementation and review.

A useful evidence structure connects requirements, decisions, responsible roles, controls, actual records, and subsequent reviews. The following examples illustrate possible implementation records. They are not prescribed document titles or a complete list of requirements.

Clause 4 / Scope: What activities does the AIMS cover?A controlled scope and context record describing organizational boundaries, relevant AI activities, interfaces, exclusions, and the basis for those boundaries. Suggested owner: AIMS owner with leadership.
Clause 5 / Authority: Who has responsibility and decision rights?Approved policy, assigned roles, decision authorities, and records showing that accountability and escalation arrangements have been used. Suggested owner: Top management and delegated owners.
Clauses 6.1.2 and 6.1.4: What risks and consequences have been assessed?Versioned risk and impact assessments documenting scope, criteria, methods, findings, limitations, and reassessment conditions. Suggested owners: Risk and impact assessment leads.
Clause 6.1.3 / Treatment: How were necessary controls selected?Documented treatment decisions, a Statement of Applicability, implementation responsibilities, management approvals, and residual-risk decisions. Suggested owners: Risk owners and designated management.
Clauses 7 and 8 / Operation: Are the arrangements actually operating?Relevant competence records, controlled procedures, assessment results, operational reviews, supplier evaluations, testing records, and change histories. Suggested owners: Process and control owners.
Clause 9.2 / Internal audit: What has independent internal examination found?An audit programme, audit criteria and scope, workpapers, findings, conclusions, and follow-up records. Suggested owner: Internal audit lead.
Clause 9.3 / Management review: What does leadership conclude from performance evidence?Management-review inputs, recorded decisions, resource needs, assigned actions, and follow-up on unresolved issues. Suggested owner: Top management.
Clause 10 / Improvement: Were identified failures actually corrected?Nonconformity records, cause analysis where appropriate, corrective actions, implementation evidence, and evaluation of effectiveness. Suggested owner: Action owner with independent verification as appropriate.

A document’s existence and a control’s operating effectiveness are different propositions. A supplier-review procedure describes what should happen. An actual review, supported by dated evidence and an accountable decision, demonstrates a different level of implementation.

Evidence should identify the relevant system, configuration, period, organizational scope, and claim it supports. Preserve material adverse findings and unresolved gaps rather than allowing favorable summary indicators to conceal them.

06 / A worked example

Can a company govern AI it does not develop?

A fictional service company introduces an externally hosted AI assistant that classifies incoming customer complaints, drafts summaries, and recommends their priority for staff review. The model and hosting infrastructure belong to a supplier, but the company controls deployment, staff workflows, and decisions about how the outputs are used.

Third-party AI implementationIllustrative example

Applying ISO/IEC 42001 to a supplier-dependent system

The supplier’s controls matter, but they do not replace the customer’s management responsibilities.

Scope and authority

The company identifies the assistant’s use within its AIMS scope, assigns a service owner and assessment responsibilities, and defines who can authorize deployment or restrict use. It distinguishes the company’s activities from those controlled by the supplier.

Risk and impact

Reviewers examine misclassification, missed urgent complaints, confidentiality, inappropriate reliance on recommendations, and the consequences of delayed service for affected customers. They examine the actual workflow, not only the model’s advertised capabilities.

Supplier dependencies

The company evaluates available supplier information and identifies gaps concerning system changes, evaluation coverage, data handling, incidents, service continuity, and responsibilities. A general supplier assurance report may help, but its scope and relevance must be examined.

Controls and applicability

The treatment process considers controls for responsible use, interested-party information, third-party relationships, relevant lifecycle activities, and additional operational safeguards. The Statement of Applicability explains the selected controls and any justified exclusions.

Operating evidence

The company documents reviewer authority, tests escalation and fallback procedures, monitors relevant error patterns, maintains supplier communication routes, and defines conditions that trigger reassessment after material changes.

Illustrative management decision

Assume the company cannot demonstrate that urgent complaints will be identified reliably or that fallback arrangements work at expected volume. Unrestricted deployment is deferred. A restricted trial may be considered only with justified controls, accountable authorization, and explicit stop conditions.

If the supplier changes the underlying model, the company must determine whether earlier evaluation and treatment decisions remain valid. If operational monitoring reveals new harmful outcomes, those findings should inform reassessment and corrective action.

The example demonstrates why third-party AI governance involves both supplier evidence and controls operated by the organization using the system. See the Vendor AI Risk guide for further treatment of supplier dependencies.

This is a fictional implementation example. The evidence, risks, and deployment decisions are illustrative and do not describe an actual certification engagement.

07 / Scope & responsibilities

Responsibility depends on what the organization does with AI.

The standard is not limited to organizations that train their own models. Procurement, integration, operation, oversight, and external provision of AI can all create relevant management responsibilities.

AI developer: Govern the development activities under organizational control.Consider design and training activities, data dependencies, evaluation, documentation, deployment, and operation as relevant to the selected scope.
AI service provider: Govern the service and its external commitments.Consider service boundaries, information supplied to customers, lifecycle processes, operational changes, and external responsibilities.
AI customer or deployer: Govern acquisition, configuration, and actual use.Evaluate supplier dependencies, internal operating arrangements, affected people, risk acceptance, human oversight, and the circumstances under which use must change or stop.
Group or multi-entity organization: Define which units, interfaces, and activities are covered.Clarify centralized and local responsibilities, shared services, external providers, and the exact management-system boundary. Do not assume all subsidiaries or uses are covered by one certificate.

Effective scoping avoids two opposite failures: defining such a broad boundary that claims cannot be substantiated, or defining a narrow boundary that overlooks dependencies necessary for responsible operation.

08 / Audit & certification

Readiness, conformity, and certification are different conclusions.

An organization can implement ISO/IEC 42001 without seeking independent certification. When certification is pursued, the conformity assessment examines the AIMS within an identified scope against the applicable requirements of the standard.

Internal readiness review: Determine what is established and what remains unsupported.Evaluate management-system arrangements, applicable requirements, records, actual operation, unresolved gaps, internal audit, and management review. A favorable readiness result is not a certificate.
Internal audit: Examine conformity and operation using appropriate objectivity.Apply an audit programme, defined criteria, competent reviewers, and records that support conclusions. Findings should drive correction and improvement where needed.
Independent certification: Obtain a conclusion from a certification body.Establish the proposed certification scope, confirm the body’s relevant competence and recognition, undergo its assessment process, and address identified nonconformities.
Certification-body requirements: ISO/IEC 42006:2025This companion standard specifies additional requirements for bodies auditing and certifying AI management systems, supplementing ISO/IEC 17021-1. It is not a second AIMS implementation standard for the organization seeking certification. [4]

When evaluating a certification claim, examine the certificate holder, exact scope, applicable edition, issuing body, accreditation or other recognition as relevant, and current validity. Avoid interpreting a scoped management-system certificate as a technical guarantee about every system or model.

Certification also does not automatically establish compliance with all applicable laws. Legal duties must be assessed according to the organization’s roles, activities, jurisdictions, and current requirements.

09 / Implementation sequence

Build an AIMS around accountable work and evidence.

The following is an InfoSecured implementation sequence for organizing the work. It is not a mandatory ISO project methodology or a substitute for the standard’s full requirements.

Define mandate, scope, and authority

Identify leadership sponsorship, the AIMS owner, relevant AI activities, interested parties, organizational boundaries, policy, and decision responsibilities.
Establish the AI-system and use-case baseline

Identify in-scope systems, uses, owners, dependencies, suppliers, relevant changes, and operating contexts. Connect records to clear identifiers instead of maintaining disconnected lists.
Establish risk and impact processes

Set assessment criteria, methods, roles, and review conditions. Perform relevant assessments and record uncertainty and limitations alongside findings.
Select, justify, and operate controls

Determine treatment needs, compare necessary controls against Annex A, record the Statement of Applicability, assign implementation responsibilities, and verify operation.
Evaluate the management system

Monitor selected indicators, conduct internal audits, and complete management reviews using evidence from actual operations and unresolved issues.
Correct gaps and maintain the AIMS

Assign corrective action, examine effectiveness, preserve historical decisions, and reassess affected controls and evidence when systems, suppliers, risks, or operating conditions materially change.

For an initial implementation, prioritize a credible organizational boundary, functioning ownership, consequential AI use cases, and the evidence most important to governance decisions. Expand the programme in a controlled way without treating an initial subset of records as full conformity.

11 / Implementation toolkit

From standard requirements to usable governance records.

InfoSecured is developing an ISO/IEC 42001 AI Governance & Implementation Toolkit to help organizations structure implementation work, maintain traceable records, and prepare evidence for internal review and audit.

The toolkit is designed to support a connected workflow rather than a collection of unrelated templates. The intended approach is to maintain controlled organizational records, link related decisions, and identify missing or contradictory evidence.

Implementation workbooks: Organize linked management-system records.Track relevant requirements, organizational responsibilities, AI systems, assessments, controls, evidence, findings, and action status using consistent record structures.
Implementation guidance: Explain how the work fits together.Provide practical instructions for scope, roles, risk and impact assessment, operational governance, review, and improvement.
Editable document templates: Support controlled organizational documents.Develop appropriate policies, management-system records, assessment reports, audit documentation, management-review records, and corrective-action materials.
Worked examples: Show what traceability looks like.Illustrate connected records using fictional scenarios, including unresolved findings and the distinction between completed documentation and effective operation.
Validation utilities: Identify defined structural and relationship issues.Support checks of workbook integrity and recorded dependencies. Automated checks cannot determine whether real evidence is sufficient or whether an organization conforms to the standard.

Release status: The implementation toolkit remains under development. A final public release and download location are not yet available. The toolkit is an independent InfoSecured resource, not an ISO or IEC publication, conformity assessment, or certification guarantee.

The Evidence Library contains currently published resources. The official ISO/IEC 42001 standard must be obtained separately through an authorized source.

12 / Common questions

Questions about ISO/IEC 42001 implementation.

What is ISO/IEC 42001?

ISO/IEC 42001:2023 is an international management-system standard for organizations developing, providing, or using AI. It establishes requirements for organizational AI governance, risk and impact management, operation, performance evaluation, and improvement.

Is ISO 42001 certification mandatory?

The standard is not universally mandated by law, and organizations can implement it without certification. Specific legal, procurement, contractual, or sector requirements may create separate obligations. Determine applicability from the relevant authority and context.

How many Annex A controls does ISO 42001 contain?

ISO/IEC 42001:2023 contains 38 reference controls across nine subject areas in Annex A. Their application depends on the organization’s control-selection and risk-treatment process. The Statement of Applicability records and justifies the relevant selections and exclusions.

What is the difference between ISO 42001 and ISO 27001?

ISO/IEC 42001 concerns AI management systems. ISO/IEC 27001 concerns information security management systems. Their governance mechanisms may overlap, but neither certification automatically establishes conformity with the other standard.

Does ISO 42001 apply if we only use third-party AI?

Yes, the standard is intended for organizations using as well as providing AI systems. The applicable management arrangements and controls depend on the organization’s role, scope, supplier dependencies, intended uses, and assessed risks.

Does certification prove that an AI model is safe or unbiased?

No. A management-system certificate concerns conformity of the AIMS within its defined scope. Specific claims about model safety, reliability, fairness, or performance require appropriate system-level evaluation and evidence.

Can ISO 42001 replace EU AI Act compliance?

No. The EU AI Act establishes legal requirements for covered activities. An AIMS may help organize some related governance processes, but the applicable regulatory duties and conformity assessment requirements must be addressed separately.

Do we need to purchase the official standard?

Organizations implementing or auditing against ISO/IEC 42001 should obtain authorized access to the standard’s requirements. Independent guides and implementation templates can assist with the work but are not substitutes for the official publication.

13 / Sources & approach

Sources & editorial approach

This guide explains the structure and practical application of ISO/IEC 42001:2023 using the published standard’s identified requirements and authoritative publication information. Implementation examples, suggested records, responsibilities, and working sequences are InfoSecured’s independent interpretations rather than official ISO templates. The guide does not reproduce the full licensed standard.

ISO/IEC 42001:2023 — Information technology — Artificial intelligence — Management systemISO publication record, edition and scope; primary standard covering the AIMS requirements and related annexes.
IEC Webstore — ISO/IEC 42001:2023Independent publisher listing confirming the edition, publication status, intended organizational users, and scope.
ISO/IEC 42005:2025 — AI system impact assessmentCompanion guidance on structured AI-system impact assessment.
ISO/IEC 42006:2025 — Requirements for bodies providing audit and certification of artificial intelligence management systemsCompanion requirements for AIMS auditing and certification bodies.
ISO/IEC 23894:2023 — Guidance on AI risk managementCompanion risk-management guidance for organizations developing, providing, deploying, or using AI.
NIST — AI Risk Management FrameworkComplementary voluntary AI risk-management framework, not an ISO conformity or certification instrument.

Source review: 17 October 2026. Fictional examples and suggested evidence structures are original explanatory guidance. Implementation and certification conclusions require evaluation against the applicable official requirements, organizational context, and actual evidence. Editorial standards.

Back to the beginning