ISO/IEC 42001:2023
The first management-system standard for artificial intelligence. Track an AI management system the same way you track your ISMS — policies, roles, impact assessment, life cycle, data governance, and third-party AI.
What ISO/IEC 42001:2023 is
ISO/IEC 42001:2023 is the first international management-system standard for artificial intelligence. It is built on the same Annex SL skeleton as ISO/IEC 27001, which means that if you already run an ISMS you already know its shape: a scope, a policy, assigned roles, a risk and impact process, documented objectives, operational controls, internal audit, and management review — but pointed at the AI systems your organisation develops, provides, or uses rather than at information assets in general.
The part most teams actually work through is Annex A, a set of controls grouped from A.2 (policies related to AI) through A.10 (third-party and customer relationships). It covers ground that no prior security standard covered: an AI system impact assessment distinct from a privacy or security assessment; data provenance and preparation as first-class controls; documentation aimed at the people affected by a system rather than at auditors; and a life cycle that treats an intended-use statement as a control rather than a marketing artefact.
What 42001 is not is a product-safety certification. It certifies that you manage AI responsibly, not that any individual model is safe, fair, or accurate. Regulators and enterprise buyers treat it the way they treat 27001 — as evidence that the discipline exists and is audited, which is exactly why it has become the fastest thing to be asked for in an AI vendor review.
In practice the work splits unevenly. The governance controls — policy, roles, reporting of concerns, resources — are familiar to anyone who has run an ISMS and are largely a documentation exercise on top of structures you already have. The life-cycle controls under A.6 are where engineering has to participate, because objectives for responsible development, verification and validation, deployment, monitoring and technical documentation are all things that have to be true of the pipeline, not of a policy document. And the data controls under A.7 — provenance, quality, preparation, acquisition — are where most organisations discover their real gap, because the answer to "where did this training data come from and under what licence" is frequently nobody's job today.
The other thing worth knowing before you start is that 42001 is deliberately scalable. The standard expects the depth of your AI management system to be proportionate to the impact of the AI systems in it, and a considered, documented decision that a control is not applicable to your context is a legitimate outcome rather than a gap. That is why an organisation that fine-tunes nothing and buys everything can still certify — and why an impact assessment process is the control everything else hangs off.
Who it applies to
- Organisations shipping a product with AI in it that are now being asked, in security questionnaires, how that AI is governed.
- Organisations that mainly *buy* AI — the obligations under A.10 (third-party and customer relationships) apply to procurement, not just to model builders.
- Anyone preparing for the EU AI Act who wants an auditable management system in place before the obligations bite. 42001 is not the Act and does not confer conformity, but it is the closest certifiable structure to what the Act asks a provider to demonstrate.
- Existing ISO 27001 certified organisations, for whom 42001 is an extension of a system that already exists rather than a new one.
How it is structured in CMDash
CMDash seeds the Annex A controls, using the published Annex A numbering — including the two-tier A.6 split between the A.6.1.x objectives and the A.6.2.x life-cycle controls, which is a real feature of the standard rather than a flattening choice. Clauses 4–10 are the management-system requirements; they are what an auditor certifies against, and they are not modelled as assessable control rows.
| Policies Related to AI | 3 |
| Internal Organization | 2 |
| Resources for AI Systems | 5 |
| Assessing Impacts of AI Systems | 4 |
| AI System Life Cycle | 9 |
| Data for AI Systems | 5 |
| Information for Interested Parties | 4 |
| Use of AI Systems | 3 |
| Third-Party and Customer Relationships | 3 |
These counts are the control rows CMDash actually seeds, checked against the seed files by a test on every build.
Scope: Annex A controls. Clauses 4–10 are the management-system requirements and are not modeled as controls.
The control text
ISO standards are sold per copy and may not be redistributed, so the control text cannot be embedded in this product.
CMDash ships the control identifiers, the titles, and its own evidence-requirement guidance. If you hold a licence, you can import the official text yourself through the framework CSV import, which records it as customer-provided.
Buy from the ISO Store →Sold per copy by the publisher
What CMDash does for ISO/IEC 42001:2023
- All Annex A controls are tracked individually, each with an evidence requirement written for collection — "RACI chart for AI roles (model owner, risk owner, reviewer, approver); appointment records" rather than a restatement of the control title.
- Seeded crosswalk mappings tie Annex A to ISO 27001:2022 Annex A and, through NIST AI RMF, to the AI RMF Core — so an access-control or supplier-management artefact you already collect for your ISMS can carry into the AIMS instead of being gathered twice.
- The regulatory radar tracks the EU AI Act as a dated entry and maps its obligations onto twelve specific ISO 42001 controls, giving a readiness percentage against the Act computed from your own control statuses rather than a generic checklist.
- AI-system impact assessments (A.5) and life-cycle documentation (A.6) get the same dated evidence and reviewer workflow as any other control, which is what makes them producible on demand a year later.
- CMDash publishes its own AI self-assessments — OWASP LLM Top 10 (2025) and CSA AI-CAIQ v1.1 across all 18 AICM domains — including the items it does not satisfy. If you are assessing us as a supplier under A.10, that is already written down.
- Every control is a tracked row with an owner, a status (not started, in progress, implemented, not applicable, at risk, failing), reviewer notes, and its own evidence requirements — not a checkbox on a spreadsheet.
- Evidence is uploaded against a specific control, dated, and kept on that control, so an assessor can see what was true when.
- Gaps export to a POA&M CSV — every control that is not implemented and not marked out of scope, with owner, status and evidence requirements.
- Policies you author in CMDash link to the controls they govern, so each control row shows which living policies stand behind it.
- A time-boxed, read-only external assessor seat can be scoped to just this framework, with an expiry date you set and a record of every page they opened.
What it shares with your other frameworks
CMDash seeds control-to-control crosswalk mappings, each with a confidence score. They are what stops the same artefact being gathered twice, and they are how a second framework costs a fraction of the first.
Where the evidence comes from
CMDash reads from the security tools you already run and turns what they report into control evidence. These are the integrations whose data lands closest to ISO/IEC 42001:2023’s control groups — you decide which control each signal evidences, and CMDash keeps it current from then on.
Where CMDash stops
Published on purpose. A buyer who discovers a limit after signing is a support ticket and a refund conversation; a buyer who reads it here is making an informed decision.
- ISO sells the text of 42001. CMDash ships control identifiers, titles and its own evidence-requirement guidance, and does not reproduce the standard's wording. You need a licensed copy of the standard to work from; there is no way around that, and a vendor who has quietly redistributed it has created a problem for you, not solved one.
- Clauses 4–10 are not modelled as controls. If you want them tracked as assessable items you would add them as custom controls.
- CMDash is not a certification body. Certification against 42001 is issued by an accredited body after an audit; what CMDash does is make that audit survivable.
Common questions
Is ISO 42001 the same thing as complying with the EU AI Act?
No, and anyone telling you otherwise is selling something. The AI Act is law with its own scope, risk tiers and conformity-assessment routes; 42001 is a voluntary management-system standard. They overlap heavily in governance, documentation and risk process, which is why CMDash maps the Act's obligations onto specific 42001 controls — but a 42001 certificate is not a presumption of conformity with the Act.
We already have ISO 27001. How much of it counts?
Structurally, most of the management system: scope, leadership, competence, internal audit and management review are the same clauses pointed at a different object. On the control side, the seeded crosswalk between 42001 Annex A and 27001:2022 Annex A is where the reuse actually is — supplier management, access control and incident response carry over well; impact assessment, data provenance and life-cycle documentation are genuinely new work.
We only use AI, we do not build it. Does 42001 apply?
Yes, in a reduced form. The standard explicitly covers organisations that provide or use AI systems, and A.9 (use of AI systems) and A.10 (third-party and customer relationships) are written for exactly that case. Your impact assessments and your supplier due diligence do most of the work; the life-cycle controls largely do not apply and would be marked not applicable with a justification, which CMDash records as a status rather than deleting the row.
How long does it take?
CMDash will not print a number it cannot compute. What it does instead is estimate a date from your own throughput — controls moved to implemented over the last 30 days against controls remaining — and show nothing at all until there is enough movement to estimate from.
Also in AI Governance
Paid plans cap how many frameworks you run at once, never which ones — ISO/IEC 42001:2023 is available on every paid plan.
View pricingReady to see your posture?
Start a 30-day free trial of any paid plan — a card is required, and it converts unless you cancel — or sign up for Free with no card at all.