ISO 42001: the 2026 guide to AI management systems.

What ISO 42001 is — and why it exists

ISO/IEC 42001:2023 is the first international standard for an Artificial Intelligence Management System (AIMS). Published in December 2023, it gives an organisation a structured, certifiable way to govern how it develops, procures and uses AI — the same way ISO 27001 gives it a way to govern information security.

It exists because AI moved faster than anyone's governance did. Boards approved tools they couldn't explain, regulators started asking pointed questions, and procurement teams began writing "how do you govern AI?" into their security questionnaires. 42001 is the answer that doesn't require inventing your own framework: an internationally recognised management system that says, in audit-ready terms, we know what AI we run, what it could do, and who is accountable for it.

Who is adopting it first

The early movers are the organisations whose customers are already asking. The large cloud and AI vendors certified quickly to take the question off the table in enterprise sales — and in doing so made it a line item their customers now have to answer too.

• Major AI & cloud providers. Microsoft, AWS and other hyperscalers pursued 42001 certification early, turning it into a trust signal in enterprise procurement.
• Enterprise procurement. "How do you govern AI?" is now appearing in vendor security questionnaires — and a 42001 certificate is the cleanest way to answer it once, not per-deal.
• Government & regulated tenders. AU and UK public-sector and regulated buyers are beginning to reference AI governance expectations, with 42001 as the recognisable benchmark.

The AI Management System — analogous to 27001's ISMS

If you already run a 27001 ISMS, the shape of 42001 will feel familiar. It uses the same Plan-Do-Check-Act lifecycle and the same harmonised management-system structure (context, leadership, planning, support, operation, evaluation, improvement). The machinery is the same; the subject is AI instead of information security.

The Annex A controls — what's there

Annex A of 42001 lists 38 controls grouped into nine objectives, with Annex B giving implementation guidance for each. They're the AI-specific work: where the management-system clauses are reused from 27001, these controls are what make it an AI management system.

AI risk management & the AI impact assessment

42001 frames AI risk on two axes that 27001 doesn't fully cover. First, the familiar one: risk to the organisation — security, continuity, reputation. Second, the one that's new to most teams: risk to individuals and society from the AI itself — bias, opacity, safety, misuse. Both feed the AI risk register the standard expects you to keep.

The AI impact assessment

The centrepiece is the AI system impact assessment (clause 6). Where a 27001 risk assessment asks "what could happen to this asset?", the 42001 impact assessment asks "what could this AI system do to the people it touches?" — and it triggers whenever you register or materially change an AI system.

• Triggered by registration. Adding a new AI system to the inventory — or changing its purpose — is what fires the assessment, so it can't be skipped.
• Structured, not freeform. It considers affected parties, potential harms, likelihood and the controls that reduce them — documented and reviewable.
• Feeds the risk treatment. The output drives which Annex A controls apply to that system, and what evidence "responsible use" requires.

Roles, responsibilities and lifecycle management

42001 is explicit that AI needs named accountability, not diffuse ownership. Someone owns the AI policy; someone owns each registered AI system; someone signs off impact assessments. The standard wants those roles defined and resourced — the absence of a clear owner is itself a finding.

It also expects AI to be managed across its whole lifecycle, not just at launch. The same system is governed from design and development, through verification and deployment, into operation and monitoring, and finally to retirement — with controls and evidence at each stage rather than a one-time approval that never gets revisited.

How 42001 interacts with 27001 and NIST

42001 was designed to sit alongside the management systems you may already run, not to replace them. It shares the harmonised Annex SL structure, so the overlap is real and reusable.

• ISO 27001. If you run a 27001 ISMS, 42001 reuses most of the management-system machinery — leadership, document control, audit, review. The Annex A controls are the genuinely new work.
• NIST AI RMF. A useful, US-origin complement — strong on risk vocabulary and function mapping. It's a framework, not a certifiable management system, so it sits next to 42001 rather than replacing it.

Certification — who certifies, and what to expect

42001 is a certifiable standard. An accredited certification body audits your AIMS and, if it conforms, issues a certificate — the same Stage 1 / Stage 2 model as 27001, followed by surveillance audits and a three-year recertification cycle.

• Stage 1. A documentation and readiness review — is the AIMS designed, scoped and documented?
• Stage 2. An operational audit — is it actually running, with impact assessments, a populated AI register and dated evidence?
• Surveillance & recert. Periodic surveillance audits keep the certificate live, with full recertification on a three-year cycle.

The accreditation ecosystem is still young — fewer bodies offer 42001 than 27001 today — but it is expanding quickly as demand rises. For most AU/UK organisations the practical first step isn't booking an auditor; it's reaching internal readiness, which is where the maturity path below comes in.

Common myths about 42001

The standard is new enough that the misconceptions are still forming. Three come up in nearly every first conversation.

The realistic path — from "we use ChatGPT" to ML2 in 6 months

Most organisations don't start at zero policy and zero AI — they start at lots of unmanaged AI and no policy. The standard doesn't grade maturity, so Cybereen adds an internal ML0–ML3 progression to track the same AIMS the way you'd track a 27001 ISMS. A realistic target is reaching ML2 in about six months.