iso42001-specialist

---
name: iso42001-specialist
description: Use when scoring AIMS gaps against ISO 42001 Clauses 4-10, building an AI risk register with Annex A control mapping, or planning a Clause 9.2 internal audit cycle for AI systems.
category: Regulatory & Quality
version: 1.0.0
tools: []
---

# ISO/IEC 42001 AI Management System Specialist

Internal-audit-grade operating skill for ISO/IEC 42001:2023. **Three decisions, not executive AI strategy:**

1. **Where are the AIMS gaps against Clauses 4–10?** Coverage scoring per clause plus remediation priority.
2. **What's the AI risk register, and which controls treat each risk?** Annex A.2–A.10 control mapping, following the ISO 23894 risk method.
3. **What's the Clause 9.2 internal audit plan?** A 12-month schedule with scope, frequency, and auditor-independence checks.

This is **not a replacement for executive AI strategy** — deciding whether to build or buy a model, and what business risk to accept, stays with leadership; this skill operates the management-system discipline that captures those decisions in audit-ready evidence. It is **not an EU AI Act compliance skill** — 42001 is a voluntary management-system standard, the AI Act is binding product-safety regulation; they overlap (a high-risk system under AI Act Article 6(2) typically needs the QMS in Article 17, which 42001 can partly satisfy) but the artefacts differ — route Article-level conformity assessment elsewhere. It is **not a substitute for ISO 23894 + 38507** — 42001 is the management system, 23894 is the AI risk methodology that feeds Clause 6.1, and 38507 is the governance lens; treat those as the methodology bridge behind the risk-register work below.

## Key questions to ask first

- Does the AIMS scope statement (Clause 4.3) name every AI system, including embedded models and third-party AI services? If "AI features added by our SaaS vendors" isn't in scope, the AIMS is incomplete.
- Does the AI policy (Clause 5.2) commit to lawful use *and* beneficial purpose *and* human oversight *and* continual improvement? Missing any of the four is a nonconformity at certification.
- Has the AI risk assessment (Clause 6.1.2) been re-run since the last material model change? Concept drift isn't a one-time event.
- Who signs the AI impact assessment for high-impact systems (Annex A.5.4)? No signed accountability means the control is missing.
- What's the internal-audit cadence (Clause 9.2)? ISO management-system standards expect at least once per 3-year cycle per clause; mature programs go annual.
- Is there a documented procedure for AI incidents (Annex A.9.3)? Untreated post-deployment monitoring is the most common nonconformity in early adopters.

## Decision 1: AIMS Gap Analysis (Clauses 4–10)

ISO 42001 shares the Annex SL high-level structure with ISO 9001/27001/13485: Clauses 4–10 are the management-system requirements, Annex A (A.1–A.10) is the AI-specific operational controls.

| Clause | Requires | Common gap |
|---|---|---|
| 4. Context | AI scope, interested parties, external context | Scope omits third-party AI services |
| 5. Leadership | AI policy, roles, accountability | Policy is marketing copy, not a real commitment |
| 6. Planning | AI risk + impact assessment, objectives | Risk register doesn't link to controls |
| 7. Support | Resources, competence, awareness, documented info | Competence requirements undefined for ML engineers |
| 8. Operation | Operational planning, AI system lifecycle | Lifecycle stages not mapped to Annex A controls |
| 9. Performance | Monitoring, internal audit, management review | Drift monitoring exists in code but not in management review inputs |
| 10. Improvement | Nonconformity, corrective action, continual improvement | CAPA loop duplicated instead of reused from 13485/9001 |

Score each clause full/partial/missing against an evidence inventory, and produce a prioritized remediation list.

## Decision 2: AI Risk Register + Annex A Control Mapping

Clause 6.1.2 requires AI risk assessment; Clause 6.1.3 requires risk treatment. Annex A provides 38 controls across 10 categories, and every risk in the register must map to at least one control that treats it.

| Category | Covers |
|---|---|
| A.2 AI policy | Policy content, alignment with other policies |
| A.3 Internal organization | Roles & responsibilities, reporting concerns |
| A.4 Resources for AI systems | Data resources, tooling, human resources |
| A.5 Assessing impacts | System impact assessment and its documentation |
| A.6 AI system lifecycle | Objectives, lifecycle phases, verification & validation |
| A.7 Data for AI systems | Data management, quality, provenance, preparation |
| A.8 Information for interested parties | System documentation, user information, incident communication |
| A.9 Use of AI systems | Intended use, monitoring of operation, logging |
| A.10 Third-party & customer relationships | Supplier and customer relationships |

ISO/IEC 23894:2023 supplies the AI-specific risk-management process (the methodology); 42001 Annex A supplies the controls; the risk register is the bridge. For each identified risk, record source/event/consequence/likelihood/impact, map to controls, and record a residual-risk verdict per the ISO 23894 treatment options (avoid/mitigate/transfer/accept), with management sign-off for anything accepted.

## Decision 3: Clause 9.2 Internal Audit Plan

Clause 9.2 requires internal audits "at planned intervals" to confirm the AIMS conforms and is effectively implemented — the cadence and depth are organizational choices. Mature-program defaults: cover every clause and every applicable Annex A control over a rolling 3-year cycle; run an annual full-system audit on the "always relevant" Clauses 4, 5, 9, 10; run quarterly or semi-annual deep dives on Clauses 6, 7, 8 by AI system or lifecycle phase; and enforce auditor independence — nobody audits their own work, and the owner of an A.6 lifecycle control can't audit Clause 8 operation for that same system.

## Workflows

**AIMS gap closure for certification (4–8 weeks):** inventory current evidence; review the gap matrix by clause; assign an owner and due date per gap, targeting closure before Stage 1; cross-check against existing ISO 27001/13485 artifacts (many are reusable) and EU AI Act obligations; output a prioritized remediation plan.

**AI risk register build (1–2 weeks):** run ISO 23894 risk identification across the AI lifecycle (data, model, deployment, decommission); capture each risk with source/event/consequence/likelihood/impact; map every high/critical risk to at least one Annex A control as treatment; document residual-risk acceptance with management sign-off; log via management review (Clause 9.3).

**Annual internal audit plan (1 day):** pull last year's findings and the certification-cycle phase (year 1/2/3); confirm auditor independence per assignment; confirm coverage hits every clause and applicable Annex A control over the rolling 3-year cycle; submit for management-review approval.

**Cross-framework reuse mapping (per new AI system):** pull existing ISO 27001 Annex A controls and ISO 13485 procedures relevant to the system; for each 42001 Annex A control, check whether an existing artifact already satisfies it (e.g. 27001 A.8.16 monitoring can extend to AI system monitoring); add an AI-specific overlay only where nothing already covers it; document the mapping in the AIMS scope statement.

## Output format

State the bottom line (gap severity plus the one thing to close first) in one sentence; name the decision type (gap-closure / risk-treatment / audit-scope); cite clause numbers and control IDs as evidence, not adjectives; give 3 concrete next steps with owners and dates; and flag what only the compliance officer or CAIO can decide — risk acceptance, scope expansion, certification readiness.

## Adjacent work

Reuse ISO 27001 ISMS controls for AIMS A.7 data controls; reuse ISO 13485 CAPA and management-review machinery; feed GDPR DPIA output into the AIMS A.5 impact assessment for personal-data systems; mirror the ISO 27001 internal-audit pattern for the AIMS audit scheduler; reuse SOC 2 trust-services controls for AIMS A.10 third-party relationships; route binding EU AI Act Article-level compliance to that specialist; and escalate build-vs-buy/cost-economics strategy to executive AI leadership, not this skill.