ISO 42001 certification shows an accredited body has audited an AI company's AI management system against ISO/IEC 42001:2023. It requires a defined scope, an AI policy, risk and impact assessments, Annex A controls, internal audit and management review, verified in a two-stage audit and valid for three years.
Enterprise questionnaires now ask how you govern AI. ISO 42001 certification answers with audited evidence, but only when its scope covers the product customers buy.
A driving licence does not prove you will never crash, but it proves an examiner watched you drive by the rules. ISO 42001 certification works the same way for AI. This guide explains what the standard requires, how the audit works, what drives cost, how to scope it and how to judge whether now is the right time.
What ISO 42001 is
ISO/IEC 42001:2023, published on 18 December 2023, sets requirements for an AI management system. That means the policies, roles and processes an organisation uses to govern AI. Like ISO/IEC 27001, it follows Plan-Do-Check-Act, so policy, risk assessment, internal audit and management review appear in both.
Two newer standards complete the picture. ISO/IEC 42005, published in May 2025, guides AI system impact assessments. ISO/IEC 42006, published in July 2025, sets requirements for the bodies that certify AI management systems.
A certificate from an accredited body is valid for three years, with annual surveillance audits. It does not certify a model's accuracy or prove legal compliance. It shows that a system for governing AI exists, is followed and improves.
What you need to build
The requirements have two parts. The main clauses apply to every certified organisation. Annex A offers 38 controls across nine areas, which you select and justify in a Statement of Applicability, the document explaining which controls apply and why.
| Area | What auditors expect | Typical evidence |
|---|---|---|
| Context and scope (clause 4) | Which AI systems and roles are covered | Scope statement and AI system inventory |
| Leadership and policy (clause 5) | An AI policy and named roles | Approved policy, role descriptions, objectives |
| Planning and risk (clause 6) | Risk and impact assessment, and risk treatment | Risk register, impact assessments, Statement of Applicability |
| Support and operation (clauses 7 and 8) | Competence and controls running in practice | Training, release and supplier records |
| Performance and improvement (clauses 9 and 10) | Monitoring, internal audit and corrective action | Audit report, review minutes, closed actions |
In our view, three Annex A areas need the most new work in an AI software company. They are impact assessment, data for AI systems, and third-party relationships, because most products depend on outside models and datasets.
How the audit works
Certification follows the familiar two-stage model. The Cloud Security Alliance describes the typical shape as follows.
- Stage 1, design review: the auditor reviews scope, policies, assessments and the Statement of Applicability, typically over one to two days.
- The gap: Stage 2 usually follows four to twelve weeks later. Beyond six months, Stage 1 may need repeating.
- Stage 2, operating effectiveness: the auditor samples real operation, typically over three to nine or more days.
- Decision and surveillance: a three-year certificate follows, with annual audits of about a third of the initial time.
Auditors need records, not intentions. Before Stage 2, run at least one real release, one internal audit and one management review. Documents written the month before, with no evidence of use, invite nonconformities.
Cost drivers and scope
There is no standard price, because cost follows scope and maturity. The main drivers are the AI systems, sites and teams in scope, and the headcount auditors must sample. An existing ISO/IEC 27001 system helps, because its processes can be reused. Your build approach and the certification body's day rates and travel complete the picture.
Ask at least two bodies to quote against an identical written scope. Confirm on the accreditation register that each is accredited for ISO/IEC 42001, not only ISO/IEC 27001. Budget for annual surveillance audits too.
A scoping example (hypothetical)
An 80-person company sells an AI contract-review product built on a third-party model. It also runs an internal coding assistant and two research prototypes. Scoping everything means more audit days and impact assessments for prototypes that may never ship. Scoping only the coding assistant is cheaper but useless in sales.
The sensible scope is the contract-review product, its data pipeline, the teams behind it and the third-party model relationship. That is what buyers will ask about, and it can widen later.
ISO 42001 does not replace security assurance. Buyers will still ask for ISO/IEC 27001 or a SOC 2 report, because a well-governed model can still run on an insecure platform. With ISO/IEC 27001 in place, adding ISO 42001 is mostly new AI content on a familiar structure.
Deciding whether to certify now
Use these signals to decide whether to pursue certification now, prepare and certify later, or wait.
| Signal | Pursue now | Prepare, certify later | Wait |
|---|---|---|---|
| Customer demand | Buyers name ISO 42001 in RFPs | AI governance questions, standard not named | No AI governance questions yet |
| AI in the product | AI is core to the offer | AI features are growing | AI is incidental or internal |
| Existing system | ISO/IEC 27001 certified | Policies exist, no certified system | No management system |
| Regulatory exposure | EU customers raise AI Act questions | Some EU exposure | None foreseeable |
| Records | Release, audit and review documented | Processes written, not yet run | Processes not yet defined |
Timing matters. EU AI Act Article 50 obligations have applied since 2 August 2026, and Annex III high-risk obligations were postponed to 2 December 2027, roughly one audit cycle away. Certification does not by itself make a company compliant, but it produces the inventories and assessments regulators ask for. Gartner also projects that AI agents will intermediate more than USD 15 trillion of B2B purchases by 2028.
The bottom line
Conclusion
A driving licence matters because an examiner watched you drive. An ISO 42001 certificate matters because an accredited auditor watched your AI management system operate, and that value depends on the right scope and real records.
Certification is rarely worth rushing, and rarely worth ignoring once buyers ask about AI governance. This week, write a one-page scope statement: the AI product customers buy, its data pipeline, the third-party models behind it and the teams that support it. Every later decision depends on that page.
Frequently asked questions
How long does ISO 42001 certification take?
It depends on scope and maturity. Once the system has produced records, Stage 2 typically follows Stage 1 by four to twelve weeks, and the gap should not exceed six months.
Does ISO 42001 replace ISO 27001 or SOC 2?
No. ISO/IEC 42001 covers AI management, while ISO/IEC 27001 and SOC 2 cover information security. The shared management system structure lets ISO 27001 and ISO 42001 run side by side.
Does ISO 42001 certification mean EU AI Act compliance?
Not by itself. A certified AI management system produces governance evidence that helps with regulatory obligations, but specific AI Act duties should be checked separately with counsel.
What is an ISO 42001 Statement of Applicability?
The document listing which Annex A controls apply to the AI management system, why each is included or excluded, and how the included controls are implemented.
How can I check whether an ISO 42001 certificate is credible?
Check that the issuing body is accredited for ISO/IEC 42001, that the certificate is in date, and that its scope names the product, entity and sites relevant to the purchase.
References
- IEC Webstore, ISO/IEC 42001:2023 Artificial intelligence management system
- ISO, ISO/IEC 27001 Information security management
- Secureframe, SOC 2 Trust Services Criteria explained
- European Commission, Transparency obligations under Article 50 of the AI Act
- Gibson Dunn, EU AI Act Omnibus: postponed high-risk deadlines
- Digital Commerce 360, Gartner: AI agents and $15 trillion in B2B purchases by 2028
- ISO, ISO/IEC 42005:2025 AI system impact assessment (May 2025)
- ISO, ISO/IEC 42006:2025 Requirements for bodies providing audit and certification of AI management systems (July 2025)
- Cloud Security Alliance, What to expect in the ISO 42001 certification process (23 July 2025)
- BLCK ALPACA, ISO 42001 Annex A controls: 38 controls across nine areas
Disclaimer: General information only, not legal, security, compliance or financial advice. Check the cited sources and seek professional advice. Names belong to their owners; no endorsement implied.