ISO/IEC 42001 — AI Management System

Published December 2023, it specifies requirements for an AI Management System (AIMS) — the first international standard for AI management systems. It is certifiable, follows the Harmonized Structure, and uses the PDCA cycle.

Annex A contains 38 normative controls. Organizations must either implement each control or provide a documented, risk-based justification for its exclusion in their Statement of Applicability (SoA). Simply ignoring a control is a nonconformity.

A two-stage third-party audit: Stage 1 reviews documentation and AIMS design; Stage 2 evaluates implementation effectiveness on-site. Certification is valid for 3 years with annual surveillance audits. Recertification requires a full audit before the certificate expires.

Both follow the Harmonized Structure (same clause numbering 4-10, same PDCA cycle). AI risks overlap with information security risks. Organizations can extend existing ISMS to include AIMS requirements, sharing processes for risk assessment, internal audit, management review, and corrective action. An integrated management system reduces duplication.

ISO 42001 (Clause 6) requires risk assessment to consider impacts on individuals, groups, and society — not just organizational/business risks like confidentiality, integrity, and availability (ISO 27001's focus). ISO 42001's risk scope includes human rights, fairness, safety, and societal impacts, which is broader than traditional information security risk assessment.

The SoA documents which of the 38 Annex A controls are implemented and which are excluded, with justification for each exclusion. It mirrors ISO 27001's SoA concept (which covers 93 controls). Both serve as a key audit artifact — auditors verify that every control is addressed. The SoA is a mandatory document for certification.

Annex A domains include: A.2 AI Policies, A.3 Internal Organization, A.4 Resources for AI Systems, A.5 Assessing Impacts of AI Systems, A.6 AI System Lifecycle, A.7 Data for AI Systems, A.8 Information for Interested Parties, A.9 Use of AI Systems, and A.10 Third-party/Supply Chain.

PDCA = Plan-Do-Check-Act. Plan: Establish AI policy and conduct risk assessments. Do: Implement Annex A controls and deploy AI systems with safeguards. Check: Monitor AI system performance, conduct internal audits, review effectiveness. Act: Address nonconformities with corrective actions and drive continual improvement. The cycle repeats continuously, ensuring the AIMS evolves with changing risks and organizational needs.