ISO/IEC 42001

• ISO/IEC 42001:2023 sets the requirements for an AI management system (AIMS): the policies, roles, risk processes, records and reviews an organisation needs to manage AI responsibly.
• It follows the same "harmonised structure" as ISO 27001 and ISO 9001: Clauses 4 to 10 are mandatory, and Annex A is a menu of 38 AI controls in 9 groups (A.2 to A.10) that you include or exclude through a Statement of Applicability (SoA).
• It is voluntary and certifiable. An accredited certification body audits you and issues a three-year certificate.
• It is a governance standard, not a technical security standard. It has no prompt-injection, jailbreak, model-abuse or runtime security controls. Pair it with NIST AI 600-1, the OWASP LLM Top 10 (2025) and MITRE ATLAS for that layer.
• Status (as of 2026-10-19): Edition 1 is current. The iso.org page shows stage 60.60 (published) with no amendment or revision project listed. Two companion standards were added in 2025: ISO/IEC 42005 (impact assessment guidance) and ISO/IEC 42006 (rules for certification bodies).

• Any organisation, of any size or sector, that provides, develops or uses AI systems. The standard asks you to work out your AI role (for example provider, producer, customer, user) as part of Clause 4.
• No legal reach. It applies only where an organisation chooses it, or where a customer, regulator or contract asks for it.
• Commonly requested in enterprise procurement, especially by EU and US buyers, and increasingly referenced as supporting evidence for EU AI Act readiness (see EU AI Act).

• No fines or legal penalties. Nobody enforces ISO 42001 by law.
• The real "enforcement" is the certification audit. Certification bodies raise major or minor nonconformities; majors must be closed before a certificate is issued or kept.
• Unresolved nonconformities can lead to suspension or withdrawal of the certificate, which can cost you deals that required it.
• Certification process:
  • Use a certification body that is accredited for ISO/IEC 42001 by a national accreditation body that is a member of the IAF (in India, check NABCB; elsewhere, for example, ANAB or UKAS). ISO itself does not certify anyone.
  • Stage 1 audit: a readiness and documentation review (scope, AI policy, risk and impact assessment method, SoA, internal audit and management review evidence).
  • Stage 2 audit: tests whether the system actually runs as documented, with interviews, samples and records.
  • Certificate is valid for three years, with surveillance audits (normally yearly) in years one and two and a recertification audit before expiry.
  • ISO/IEC 42006:2025 now sets the extra competence and audit-time rules that AIMS certification bodies must meet, on top of the general ISO/IEC 17021-1 rules.

• SaaS and AI product exporters: ISO 42001 is fast becoming the AI equivalent of ISO 27001 in security questionnaires. A certificate shortens enterprise sales cycles in the EU, UK and US. If you already hold ISO 27001, roughly half the management system (Clauses 4 to 10 shape, document control, internal audit, management review) can be reused.
• BFSI (banks, NBFCs, payment firms): RBI's RBI FREE-AI recommendations (board-approved AI policy, risk tiers, inventory, model lifecycle governance, audit) line up closely with Clauses 5, 6 and 8 and Annex A.2, A.5, A.6. An AIMS is a practical way to show RBI supervisors structured governance, though it does not cover FREE-AI's red teaming, kill switch or customer-facing items on its own.
• IT services and GCCs: clients will push ISO 42001 duties down through contracts (Annex A.10). Expect requests for your AI inventory, impact assessments and supplier controls.
• Personal data: ISO 42001 does not replace DPDP Act compliance. Map your lawful basis and retention work alongside Annex A.7.
• Security gap: if your product is an LLM app or agent, buyers will still ask how you defend against prompt injection and data leakage. Certification does not answer that; SC-AI-029 AI-specific security controls (prompt injection, model abuse) and SC-AI-018 Robustness and adversarial testing (red teaming) do.
• Budget for: a purchased copy of the standard, a gap assessment, one internal audit cycle, a management review and Stage 1 plus Stage 2 audit fees.

Clause rows are mandatory for certification. Annex A rows summarise each control group; individual controls apply only if selected in your SoA. All wording is a paraphrase.

• AIMS scope statement and the list of in-scope AI systems.
• Context analysis: interested parties register, legal and regulatory register, your AI role per system.
• Approved AI policy with evidence of communication (intranet page, acknowledgements).
• RACI or role descriptions for the AIMS, including the person who reports to top management.
• AI risk assessment method, risk criteria, risk register and treatment plan with owner sign-off.
• Statement of Applicability listing all 38 Annex A controls, included or excluded, with reasons.
• AI system impact assessments (method plus completed assessments); ISO/IEC 42005 is a useful guide.
• AI objectives with metrics and progress tracking.
• Training and competence records; awareness campaign material.
• Document control procedure and version history.
• Lifecycle records per AI system: requirements, design notes, validation results, deployment approvals, monitoring dashboards, event logs.
• Data records: sources, acquisition terms, quality checks, provenance.
• Supplier and customer agreements showing AI responsibilities.
• Internal audit programme and reports; management review minutes; nonconformity and corrective action log.

• Treating ISO 42001 certification as proof an AI product is secure. It has no technical controls for prompt injection, jailbreaks, data exfiltration via the model or agent abuse.
• Swapping 10.1 and 10.2. In the 2023 edition 10.1 is continual improvement and 10.2 is nonconformity and corrective action.
• Citing A.3.3 as "training". A.3.3 is reporting of concerns; competence and awareness sit in 7.2 and 7.3, and human resources for AI in A.4.6.
• Inventing Clause 8 sub-numbers beyond 8.4, or placing change planning anywhere other than 6.3.
• Running the impact assessment (6.1.4, 8.4) as a copy of the risk assessment. The risk assessment is about the organisation; the impact assessment is about people and society.
• An SoA that excludes controls without a reason, or that silently drops A.10 because "we only use vendor AI". Vendor AI is exactly what A.10 covers.
• Quoting the standard's text in public material. It is copyrighted.
• Picking a certification body that is not accredited for ISO/IEC 42001 specifically. An unaccredited certificate carries little weight with buyers.
• Scoping the AIMS so narrowly (one chatbot) that the certificate impresses no customer.

• AIMS (AI management system): the set of policies, roles, processes and records an organisation uses to manage AI.
• Harmonised structure: the common clause layout (4 to 10) shared by ISO management system standards, so ISO 27001, 9001 and 42001 fit together.
• Annex A: the normative list of 38 AI controls in 9 groups (A.2 to A.10).
• Statement of Applicability (SoA): the document listing every Annex A control, whether you use it and why.
• AI system impact assessment: a structured look at how an AI system may affect people and society; required by 6.1.4 and 8.4.
• Nonconformity: a failure to meet a requirement; "major" findings block certification.
• Accredited certification body: an auditing firm checked by a national accreditation body (for example NABCB in India) as competent to certify against this standard.
• Stage 1 / Stage 2 audit: the two parts of the first certification audit; documents and readiness first, then real operation.
• Surveillance audit: the smaller yearly check that keeps a certificate valid between full audits.
• JTC 1/SC 42: the joint ISO and IEC committee that writes AI standards.

• Voluntary, certifiable, global: the AI counterpart of ISO 27001, published 18 Dec 2023 and still on Edition 1 with no amendment listed as of 2026-10-19.
• 27 mandatory clause obligations (4.1 to 10.2) plus 9 Annex A control groups (38 controls, chosen through the SoA): 36 rows here.
• Certification: accredited body, Stage 1 then Stage 2, three-year certificate with yearly surveillance; ISO/IEC 42006:2025 now governs the certifiers.
• Governance only: pair it with NIST AI 600-1, OWASP LLM Top 10 (2025) and MITRE ATLAS for technical AI security.
• For Indian BFSI it is a strong backbone for RBI FREE-AI; for exporters it is a sales enabler.
• The date is the publication date, not an enforcement date, and the requirement count depends on granularity (this note uses 36).

• ISO/IEC 42001:2023 AI management systems, iso.org catalogue page (status, edition, life cycle; full text must be purchased), accessed 2026-10-19
• ISO/IEC 42005:2025 AI system impact assessment, iso.org, accessed 2026-10-19
• ISO/IEC 42006:2025 Requirements for AIMS audit and certification bodies, iso.org, accessed 2026-10-19
• Clause and Annex A numbering follows the structure in iso 42001 overview; the full text was not accessed for this note, so a reviewer should confirm row references against a purchased copy. No ISO text reproduced.