Skip to content

ISO/IEC 42001

In one sentence

The world's first AI Management System standard — same shape as ISO/IEC 27001 (information security), but for AI. Certifiable, with a mandatory Statement of Applicability.

Who in your org owns it

  • CAIO — sponsor.
  • Compliance — owns certification readiness.
  • 2L risk function — runs the AIMS day-to-day.
  • Internal audit (3L) — independent assurance.

The two-minute explainer

ISO/IEC 42001:2023 was published in December 2023 and adopted as EN ISO/IEC 42001:2026 in Europe. It defines an Artificial Intelligence Management System (AIMS) — a documented, repeatable, auditable program for managing AI in an organisation.

Three things distinguish it from NIST AI RMF:

  • It's certifiable. Accredited bodies can audit a firm against the standard and issue a certificate. NIST AI RMF isn't certifiable.
  • It's management-system shaped. Same plan-do-check-act backbone as ISO 27001 (security), ISO 9001 (quality), ISO 14001 (environmental). If a firm already runs an ISO management system, 42001 plugs in.
  • It requires a Statement of Applicability (SoA) — a documented row- per-control table showing which Annex A controls are in scope and how they're implemented.

42001 has 9 Annex A control objectives covering ~38 controls:

  • A.2 — Policies related to AI
  • A.3 — Internal organization
  • A.4 — Resources for AI systems
  • A.5 — Assessing impacts of AI systems
  • A.6 — AI system life cycle
  • A.7 — Data for AI systems
  • A.8 — Information for interested parties
  • A.9 — Use of AI systems
  • A.10 — Third-party and customer relationships

What Regulus does for you

The Iso42001Framework class binds Regulus mechanisms to specific Annex A controls. Highlights:

  • A.2.2 (AI policy) ↔ ComplianceProfile + GovernanceFramework as policy-in-code.
  • A.3.2 (AI roles and responsibilities) ↔ smf_holder attribution per audit event.
  • A.4.4 (tooling resources) ↔ ModelRegistry.
  • A.5.2 (impact assessment) ↔ model-risk-tier.
  • A.6.2.7 (event logs) ↔ RegulusAuditPlugin.
  • A.6.2.5 (operation and monitoring) ↔ post-market-monitoring.
  • A.8.5 (information for users) ↔ provenance fields per event.
  • A.9.2 (intended use) ↔ purpose-binding.
  • A.10.3 (third-party suppliers) ↔ model registry + third-party-risk.

Statement of Applicability generator

StatementOfApplicability is generated from GovernanceProgramState.statusByControlId, producing one row per Annex A control with status (IMPLEMENTED / PARTIAL / MAPPED_NOT_IMPLEMENTED / NOT_APPLICABLE / GAP), justification, and evidence reference. The artefact is what a certification body asks for first.

Saves you ~

  • ~3 program-weeks to build an SoA from raw audit data without Regulus.
  • ~6 program-weeks to evidence Annex A control coverage from scratch.
  • Ongoing: SoA refresh per cycle (typically annually + on material change).

Activation

regulus:
  governance:
    frameworks: [iso-42001]

Pair with iso-23894 (risk management) and iso-23053 (AI/ML architecture) to complete the ISO governance stack.

What an assessor will ask

  1. "Show me the SoA." Generated artefact + signed snapshot.
  2. "How does control A.6.2.7 operate?" Walk an audit event from the topic to the GRC tool.
  3. "How do you handle change management for AI systems?" Demonstrate the kill switch + model registry version pinning + audit linkage.
  4. "Where is policy documented?" ComplianceProfile + GovernanceFramework declarations point to the policy text.

What this doesn't cover

  • The certification audit itself. External body; you commission it.
  • AIMS process documentation beyond what Regulus encodes (organisational charts, training records, internal reviews).
  • Annex B and Annex C guidance. Implementation-supporting only.
  • AIMS scope definition. Your firm decides what's in scope; Regulus exposes the artefact.

Citations

  • ISO/IEC 42001:2023 — https://www.iso.org/standard/42001
  • EN ISO/IEC 42001:2026 (European adoption).
  • ISO 42001 Annex A — published with the standard.