ISO/IEC 42001:2023 Audit-Readiness Rubric — Artificial Intelligence Management System (AIMS)
What this is: ISO/IEC 42001:2023, "Information technology — Artificial intelligence — Management system," published 18 December 2023 by ISO/IEC JTC 1/SC 42, is the world's first certifiable management-system standard for AI — it specifies requirements for organizations that develop, provide, or use AI systems, covering the whole AI lifecycle rather than a single product or model. It is built on ISO's Annex SL harmonized high-level structure — the same skeleton ISO/IEC 27001 uses — so it is deliberately structured as ISO 27001's sibling: same clause numbering (4–10), same "core clauses + Annex A control catalogue + Statement of Applicability" mechanic, adapted from information-security management to AI-system governance and risk management. Relevant to Regscribe both because pharma/biotech customers are increasingly deploying AI (R&D, manufacturing analytics, pharmacovigilance signal detection) and because Regscribe itself is an AI tool operating in this exact governance space.
Sourcing note: the full ISO/IEC 42001:2023 standard text is copyrighted and sold by ISO — no verbatim clause or control text is used below. Structure and paraphrase are sourced from secondary reference sites this session: ISMS.online (isms.online/iso-42001/annex-a-controls/ — Annex A theme/control breakdown), Hicomply (hicomply.com/hub/the-core-requirements-of-iso-42001-clauses-4-10 — clause 4–10 requirement summaries), Mindsetcyber (mindsetcyber.com.au/iso-42001-controls-list/ — full 38-control title list with sub-numbering, used to cross-check ISMS.online's counts), and a general web search cross-checking publication date, JTC 1/SC 42 authorship, and Annex SL structure against ISO.org's own standard listing (iso.org/standard/42001), ANAB/ANSI's blog, and AI Standards Hub (aistandardshub.org). Where ISMS.online and Mindsetcyber's independently-published control lists agreed on control numbers/titles, that is treated as a reasonably reliable secondary-source cross-check — not a first-party ISO PDF extraction. No quantified failure-pattern or enforcement data was researched for this standard this session (mirroring the iso27001-rubric.md convention) — the "common failure pattern" column is omitted rather than fabricated; only requirement/citation/evidence are populated. Certification body accreditation timelines, exact Annex B/C/D content, and any 2025+ amendments were not independently verified this session — see Gaps below.
Rubric legend
Each row: Requirement (plain language) → Citation → Evidence that demonstrates compliance
AIMS core clauses (4–10)
| # | Requirement | Citation | Evidence of compliance |
|---|---|---|---|
| 1 | Internal/external issues affecting the AI Management System (AIMS) identified; interested parties (including those affected by AI system outputs, not only direct users) and their requirements documented; AIMS scope defined across the AI lifecycle the organization is responsible for (design through decommissioning, or the relevant subset if the org is a downstream deployer rather than a developer) | ISO/IEC 42001:2023, Clause 4 | AIMS scope statement identifying which AI systems/roles (provider, developer, deployer/user) are in scope; documented context and stakeholder analysis, including affected non-user parties |
| 2 | Top management demonstrates visible ownership: establishes and communicates an AI policy, assigns AI-related roles/responsibilities/authorities, ensures resources are provided, and promotes a culture of responsible AI development/use | ISO/IEC 42001:2023, Clause 5 | Approved AI Policy; documented role assignments (including any AI governance committee/owner role); resourcing decisions; evidence of policy communication |
| 3 | Formal AI risk assessment methodology applied covering AI-specific risk categories (bias, fairness, explainability, robustness, safety, model drift/degradation, misuse); risk treatment plan produced; Statement of Applicability documents which of the Annex A controls are implemented/excluded and why; measurable AI objectives set | ISO/IEC 42001:2023, Clause 6 | AI risk assessment methodology document; AI risk register naming AI-specific risk categories, not a generic IT-risk register reused unchanged; risk treatment plan; Statement of Applicability; documented AI management objectives |
| 4 | Resources, competence, awareness, communication, and control of documented information all addressed for the AIMS specifically (not assumed to be inherited wholesale from an existing ISMS) | ISO/IEC 42001:2023, Clause 7 | AI-specific training/competence records (data science, AI ethics, AI risk literacy — distinct from general infosec awareness training); AI-awareness program evidence; document control procedure covering AIMS records |
| 5 | Operational planning and control applied across the AI system lifecycle; AI system impact assessments performed for systems with meaningful effect on individuals, groups, or society; change management applied to model/dataset/algorithm changes; AI-specific incident response defined | ISO/IEC 42001:2023, Clause 8 | Operational procedures per lifecycle stage; documented AI impact assessments for in-scope systems; change-control records covering model/data/algorithm changes distinctly from ordinary software changes; AI incident log and response procedure |
| 6 | Monitoring/measurement/analysis/evaluation of the AIMS and of AI system performance against defined governance indicators; scheduled internal audit program; periodic management review | ISO/IEC 42001:2023, Clause 9 | AI performance/governance monitoring records; internal audit programme and reports covering the AIMS; management review minutes with defined agenda coverage |
| 7 | Nonconformities handled with root-cause analysis and corrective action; continual improvement of the AIMS demonstrated, incorporating lessons from AI incidents, audits, and stakeholder feedback | ISO/IEC 42001:2023, Clause 10 | Nonconformity/corrective action log; evidence of AIMS changes resulting from findings, incidents, or feedback |
Annex A control coverage (38 controls, 9 themes — A.2 through A.10)
Annex A is a control catalogue, not a mandatory checklist — which controls apply is determined by the organization's own AI risk/impact assessment and recorded in the Statement of Applicability, exactly as with ISO 27001 Annex A. Control-count and title breakdown below cross-checked against two independent secondary sources this session.
| # | Requirement | Citation | Evidence of compliance |
|---|---|---|---|
| 8 | Policies related to AI documented, aligned with other organizational policies, and periodically reviewed (3 controls: A.2.2 AI policy, A.2.3 alignment with other organisational policies, A.2.4 review of the AI policy) | ISO/IEC 42001:2023, Annex A §A.2 | Approved AI policy document; cross-reference showing alignment with adjacent policies (e.g. information security, data protection, quality); policy review schedule and records |
| 9 | AI-specific roles/responsibilities defined and a channel exists for reporting concerns about AI systems (2 controls: A.3.2 AI roles and responsibilities, A.3.3 reporting of concerns) | ISO/IEC 42001:2023, Annex A §A.3 | RACI or equivalent naming AI-specific roles; documented concern-reporting channel (internal or external-facing) and evidence it is monitored/acted on |
| 10 | Resources needed to build/operate AI systems are documented and managed across five resource categories (5 controls: A.4.2 resource documentation, A.4.3 data resources, A.4.4 tooling resources, A.4.5 system/computing resources, A.4.6 human resources) | ISO/IEC 42001:2023, Annex A §A.4 | Resource inventory/documentation per category; evidence resourcing decisions are traceable to AI system requirements |
| 11 | AI system impact assessment process defined, documented, and applied to assess impact on individuals/groups and on society (4 controls: A.5.2 impact assessment process, A.5.3 documentation of impact assessments, A.5.4 assessing impact on individuals/groups, A.5.5 assessing societal impacts) | ISO/IEC 42001:2023, Annex A §A.5 | Documented impact assessment methodology; completed impact assessment records for in-scope AI systems, covering both individual/group and societal impact dimensions separately |
| 12 | Responsible development objectives and processes defined; AI system requirements specified, designed, verified/validated, deployed, operated/monitored, technically documented, and event-logged across the full lifecycle (9 controls, with internal sub-numbering A.6.1.2–A.6.1.3 for objectives/processes and A.6.2.2–A.6.2.8 for the lifecycle stages themselves) | ISO/IEC 42001:2023, Annex A §A.6 | Documented responsible-AI development objectives; design/development records; verification and validation test evidence; deployment records; operational monitoring evidence; technical documentation per system; event/audit log configuration and retention evidence |
| 13 | Data used for AI development/enhancement is acquired, quality-assured, provenance-tracked, and prepared under documented controls (5 controls: A.7.2 data for development/enhancement, A.7.3 acquisition of data, A.7.4 quality of data, A.7.5 data provenance, A.7.6 data preparation) | ISO/IEC 42001:2023, Annex A §A.7 | Data acquisition records; data quality criteria and check records; data provenance documentation/lineage; data preparation (cleaning/labelling/transformation) procedure and records |
| 14 | Information about AI systems is documented for users, external reporting obligations are met, incidents are communicated, and information needs of other interested parties are addressed (4 controls: A.8.2 system documentation/information for users, A.8.3 external reporting, A.8.4 communication of incidents, A.8.5 information for interested parties) | ISO/IEC 42001:2023, Annex A §A.8 | User-facing system documentation; external reporting records (regulatory or voluntary, as applicable); incident-communication records/templates; documented approach to informing other interested parties (e.g. affected non-users) |
| 15 | Responsible-use processes and objectives defined for AI systems the organization uses (as a deployer, not only as developer); intended use documented and use restricted accordingly (3 controls: A.9.2 processes for responsible use, A.9.3 objectives for responsible use, A.9.4 intended use of the AI system) | ISO/IEC 42001:2023, Annex A §A.9 | Documented responsible-use procedures for AI systems the organization deploys/consumes (e.g. third-party AI tools used internally); intended-use statements; evidence use is monitored against intended use |
| 16 | Responsibilities allocated between the organization and third parties/suppliers/customers across the AI supply chain, including supplier assessment and customer-facing obligations (3 controls: A.10.2 allocation of responsibilities, A.10.3 suppliers, A.10.4 customers) | ISO/IEC 42001:2023, Annex A §A.10 | Documented responsibility-allocation matrix (own org vs. AI supplier vs. customer/deployer); supplier assessment records for AI component/model/data suppliers; customer-facing AI documentation or contractual terms |
How ISO 42001 relates to Regscribe's other rubrics
- ISO/IEC 27001 (
docs/iso27001-rubric.md): 42001 is 27001's structural sibling, not its substitute. Both use ISO's Annex SL harmonized high-level structure (clauses 4–10 map to the same seven headings — context, leadership, planning, support, operation, performance evaluation, improvement), both pair core clauses with an Annex A control catalogue scored via a Statement of Applicability, and both are third-party certifiable. The substance is complementary and non-overlapping: 27001 governs the confidentiality/integrity/availability of information (data/systems security generally); 42001 governs the responsible development, provision, and use of AI systems — bias, explainability, AI-specific risk, AI lifecycle management, AI supply chain. An organization can be 27001-certified and still have no AI governance at all, and vice versa; a mature AI-using organization should expect to need both, run as parallel-but-linked management systems (many organizations point their 42001 risk register and SoA at the same governance committee/document-control system as their 27001 ISMS, but the control substance does not overlap one-for-one). - EU GMP Annex 11 (
docs/eu-annex11-rubric.md): confirmed relevant per that rubric's own "READ THIS FIRST" section — the European Commission's draft Annex 11 revision (consultation closed 7 October 2025, final publication expected mid-2026 per that rubric's sourcing) adds a new Annex 22 (Artificial Intelligence), the first EU GMP annex to formally address AI/ML systems in GxP-regulated activities. That rubric already flags Annex 22 as "a strong candidate for its own standalone rubric entry." ISO 42001 and the prospective Annex 22 are not the same instrument — 42001 is a voluntary, certifiable, general-purpose AI management-system standard; Annex 22 will be a binding EU GMP requirement scoped specifically to AI/ML systems used in GMP-regulated pharmaceutical manufacturing/QC activities. A pharma/biotech company would likely use 42001 as the broader organizational AI-governance backbone and Annex 22 (once finalized) as the GMP-specific delta layered on top — the same relationship this codebase already models between ISO 27001 (broad) and NIS2 (binding, sector-specific) indocs/nis2-rubric.md.
Pharma/biotech relevance
ISO 42001 is generic-industry, not pharma-specific — no source found this session claims a pharma-specific version or carve-out. Its relevance to Regscribe's customer base is structural: pharma/biotech companies are increasingly deploying AI across R&D (target identification, molecule design), manufacturing analytics (process monitoring, predictive maintenance), and pharmacovigilance (signal detection in adverse-event data) — all activities where an AI system's bias, drift, or unexplainable output could have direct product-quality or patient-safety consequences, distinct from (but adjacent to) GxP concerns. It is also directly relevant to Regscribe's own operation: Regscribe is itself an AI tool performing AI-assisted document review/compliance analysis for regulated customers, meaning Regscribe's own AI governance posture (data provenance for what the model is trained/fine-tuned/prompted on, impact assessment of its outputs on customer compliance decisions, incident handling if it produces an incorrect compliance determination) is a live, first-party ISO 42001-shaped question, not only a rubric to sell.
Documents/evidence typically needed
AI Policy; AIMS scope statement; AI-specific risk assessment methodology and risk register; risk treatment plan; Statement of Applicability; AI management objectives; AI-specific competence/training records; AI system impact assessments; data acquisition/quality/provenance/preparation records; AI system technical documentation and event logs; AI incident log and response records; supplier/third-party AI responsibility-allocation records; internal audit programme and reports; management review records; nonconformity and corrective action records.
Gaps/follow-ups
- No quantified real-world audit-finding or certification-failure data was found or used for this standard this session — as with
iso27001-rubric.md, the "common failure pattern" column present in the GxP rubrics is deliberately omitted here rather than fabricated. Given ISO 42001 was only published December 2023, a mature enforcement/audit-finding dataset likely does not yet exist industry-wide; this should be re-checked in a future session rather than assumed permanently absent. - Annex A control titles and counts (38 controls, 9 themes) were cross-checked between two independent secondary sources (ISMS.online and Mindsetcyber) that agreed on numbering and titles — this is a reasonable secondary-source cross-check but not a first-party extraction from the ISO PDF itself (which is paywalled). If this rubric is used for anything with legal/audit consequence, recommend purchasing the primary ISO/IEC 42001:2023 text (and its normative Annex A / informative Annex B) to confirm word-for-word control wording before scoring against it.
- Annex B (implementation guidance), Annex C (AI-related organizational objectives/risk sources), and Annex D (sector/domain application notes) were mentioned in passing by one source (ISMS.online, via the initial search summary) but their content was not independently researched this session — flagged as unverified rather than described.
- Relationship to the EU AI Act was not researched this session and is deliberately not asserted here — ISO 42001 is often marketed as "helpful evidence toward" EU AI Act conformity by certification bodies and consultancies, but whether/how it maps to specific EU AI Act obligations (e.g. Article 9 risk management system, Article 17 quality management system for high-risk AI providers) was not verified against a primary EU AI Act source this session. Do not present ISO 42001 certification as EU AI Act compliance in customer-facing material without a dedicated research pass.
- Certification body accreditation landscape (which bodies are currently accredited to certify against ISO 42001, typical Stage 1/Stage 2 audit timelines) was not researched this session — unlike the
iso27001-rubric.md/docs/iso27001-checklist.mdpairing, which has UKAS-accredited-body-sourced certification timing detail, no equivalent research was done for ISO 42001 this session. - This rubric was drafted from secondary sources only, in a single research session on 2026-08-18 — no primary ISO.org PDF was fetched (ISO does not offer free full-text access to the standard). Treat clause/control numbers as reliable (cross-checked across sources) but treat any specific requirement wording above as paraphrase, not quotation, consistent with row 1 of the sourcing note.