ISO/IEC 42001:2023, published in December 2023, is the first internationally recognised certifiable standard for AI management systems. It specifies requirements for establishing, implementing, maintaining, and continually improving an Artificial Intelligence Management System (AIMS) within the context of an organisation. Certification is granted by accredited third-party certification bodies following a two-stage audit process.

Unlike NIST AI RMF (which is a voluntary framework with no certification path), ISO/IEC 42001 certification is a formal statement that an accredited auditor has verified your AIMS meets the standard's requirements. Enterprise procurement teams and regulated-sector customers are increasingly requiring 42001 certification as a supplier prerequisite.

This checklist covers what an auditor will look for at each clause during a Stage 1 (documentation review) and Stage 2 (implementation audit). Items marked with ⚠ are among the most commonly cited nonconformities in first-time certifications.

The Certification Path

  1. Gap assessment (internal or with a consultant): Compare your current state against all clauses 4–10 and Annex A. Produce a gap register.
  2. Remediation: Address gaps. This is the longest phase—typically 3–9 months depending on starting maturity.
  3. Internal audit: Clause 9.2 requires at least one completed internal audit cycle before Stage 2. This cannot be skipped.
  4. Management review: Clause 9.3 requires a completed management review with documented outputs.
  5. Stage 1 audit (certification body): Documentation review. The auditor reviews your AIMS documentation without requiring live evidence of operation. Any major nonconformities here must be closed before Stage 2.
  6. Stage 2 audit (certification body): Implementation audit. The auditor verifies the AIMS is operational—interviews staff, reviews records, checks that policies are being followed in practice.
  7. Certification decision: If no major nonconformities remain, the certification body issues an ISO/IEC 42001:2023 certificate. Minor nonconformities may be accepted with a corrective action plan.
  8. Surveillance audits: Annual. Certification is valid for 3 years subject to annual surveillance; recertification audit at year 3.

Clause 4: Context of the Organisation

What auditors check:

  • ☐ A documented analysis of internal and external factors relevant to the organisation's AI activities (Clause 4.1).
  • ☐ ⚠ A documented register of interested parties (Clause 4.2): who is affected by the organisation's AI systems? This must include regulators, customers, users, employees, and affected third parties—not just internal stakeholders.
  • ☐ A defined scope for the AIMS (Clause 4.3): which AI systems and organisational units are in scope? The scope must be specific enough to be auditable. "All AI systems" is usually acceptable if there is a system inventory to back it up.
  • ☐ ⚠ An AI policy that is appropriate to the scope and signed by top management (Clause 4.4 + 5.2). This is a separate document from an information security or data protection policy.

Clause 5: Leadership

What auditors check:

  • ☐ Evidence of top management commitment: the AI policy must be approved at board or C-suite level, not just by the compliance or engineering team (Clause 5.1).
  • ☐ Defined roles and responsibilities for the AIMS (Clause 5.3): who owns the AIMS overall? Who is responsible for individual AI systems? This should be documented in an RACI or equivalent.
  • ☐ ⚠ The AI management system owner must have sufficient authority and resources. Auditors look for evidence that the AIMS owner has a direct reporting line to leadership and can escalate AI risks for decision.

Clause 6: Planning

What auditors check:

  • ☐ ⚠ An AI risk and opportunity assessment process (Clause 6.1): documented methodology for identifying and assessing AI-specific risks and opportunities. This is separate from information security risk assessment.
  • ☐ Documented AI system impact assessments (Clause 6.1 + Annex A.6): for AI systems that may affect individuals, a structured impact assessment must be completed before deployment.
  • ☐ AIMS objectives with measurable targets (Clause 6.2): at least 2–3 measurable objectives for the AIMS (e.g., "100% of new AI systems to have a completed impact assessment before deployment," "zero high-severity open audit findings at any point"). Objectives must have owners, timelines, and defined measurement methods.
  • ☐ Planning for changes to the AIMS (Clause 6.3): a documented process for managing changes to AI systems that could affect the AIMS.

Clause 7: Support

What auditors check:

  • ☐ Evidence that adequate resources (people, tools, budget) have been allocated to the AIMS (Clause 7.1).
  • ☐ ⚠ Competence records (Clause 7.2): for all persons whose work affects AI system performance or the AIMS, there must be documented evidence of competence—qualifications, training completion records, or experiential evidence. This is one of the most common Stage 2 nonconformities. A training policy is not sufficient; individual competence records are required.
  • ☐ Awareness programme evidence (Clause 7.3): records showing that relevant staff are aware of the AI policy, their roles in the AIMS, and the consequences of non-conformance.
  • ☐ ⚠ Documented information controls (Clause 7.5): the AIMS must define which documents require version control, review cycles, and approval. All AIMS documents must have a document control header (title, version, date, approver). Missing or informal document control is a very common nonconformity.

Clause 8: Operation

This is the most implementation-heavy clause and the one most auditors spend the most time on during Stage 2.

  • ☐ Operational planning and control (Clause 8.1): documented processes for the AI system development lifecycle, including design review gates, testing requirements, and deployment approval.
  • ☐ ⚠ AI system impact assessments (Clause 8.4, cross-referenced to Annex A.6): completed impact assessments for all in-scope AI systems. An auditor will sample specific systems and request to see their impact assessment. "We do this informally" is not acceptable.
  • ☐ Data governance controls (Annex A.7): documented data provenance, data quality controls, and bias analysis for training data. Cross-references to EU AI Act Article 10 requirements for teams that are also EU AI Act-compliant.
  • ☐ Human oversight controls (Annex A.9): documented human oversight mechanisms for AI systems where the system's output could affect individuals. Evidence that these mechanisms are operational, not just described in documentation.
  • ☐ Third-party AI system controls (Clause 8.5): if the organisation uses third-party AI systems or models, there must be a process for assessing and managing the associated risks. Supplier AI risk assessments are an increasingly expected artefact here.

Clause 9: Performance Evaluation

  • ☐ Monitoring and measurement plan (Clause 9.1): defined metrics for AIMS effectiveness, with a documented monitoring programme. At minimum: AI system performance metrics and AIMS objective tracking.
  • ☐ ⚠ Completed internal audit programme (Clause 9.2): at least one full internal audit cycle covering all clauses must be completed and documented before Stage 2. The internal audit must be conducted by persons who are independent of the area being audited. Many first-time certifications are delayed because the internal audit has not been completed or is not sufficiently independent.
  • ☐ Management review records (Clause 9.3): documented minutes or outputs of a management review that covers the mandatory agenda items: AIMS performance, audit results, nonconformities, resource adequacy, objectives review, and external changes (regulatory, stakeholder).

Clause 10: Improvement

  • ☐ Nonconformity and corrective action process (Clause 10.1): a documented process for recording nonconformities (from internal audit, incidents, or external audit), conducting root cause analysis, and implementing corrective actions with defined timelines.
  • ☐ Continual improvement evidence (Clause 10.2): evidence that the AIMS is improving over time, not just maintaining compliance. Auditors look for closed nonconformities with verified effectiveness, objective trend data, and documented lessons learned from AI incidents.

Annex A: AI-Specific Controls

ISO/IEC 42001 Annex A contains AI-specific controls that organisations select based on their risk assessment. Key controls frequently examined in certification audits include:

  • A.6 — AI system impact assessment: Process for assessing the potential impacts of AI systems on individuals and society before deployment and when material changes occur.
  • A.7 — Data for AI systems: Controls for data quality, provenance, representativeness, and bias detection in training, validation, and test datasets.
  • A.8 — Information for interested parties about AI system use: Transparency documentation for users and affected parties about how AI systems work and what their limitations are.
  • A.9 — Human oversight of AI systems: Controls ensuring that human oversight is designed into AI system operation, with defined escalation and override mechanisms.
  • A.10 — Testing of AI systems: Structured testing requirements including bias testing, robustness testing, and adversarial testing for high-impact AI systems.

Typical Certification Timeline

Phase Typical Duration Notes
Gap assessment 2–4 weeks Faster with existing ISO 27001 programme
Remediation and implementation 3–9 months Depends heavily on starting maturity and number of in-scope AI systems
Internal audit cycle 4–6 weeks Must be completed before Stage 2; cannot be compressed
Stage 1 audit 1–2 days on-site or remote Documentation review; typically 4–8 week lead time to book
Stage 2 audit 2–5 days on-site Duration scales with scope size; must occur within 6 months of Stage 1
Certification decision 2–4 weeks post-audit Certification body review; minor nonconformity closure may extend this

Total from starting gap assessment to certificate: typically 6–12 months. Organisations with a mature ISO 27001 programme can compress this to 4–6 months by leveraging existing document control, internal audit, and management review infrastructure.