ISO 42001 certification

ISO/IEC 42001:2023 is the management system standard for artificial intelligence, and it is early. Over the 365 days to 11 August 2026 it appears in 18 EU public tender notices on TED, against 3,408 for ISO 27001. The honest reason to certify in 2026 is that a customer, an investor or a procurement list is asking.

It is not a legal deadline, and it is not a route to EU AI Act conformity. If you already hold ISO 27001, the second standard costs a fraction of the first, because clauses 4 to 10 are shared and what you add is 38 controls, one new kind of assessment, and the records that go with them. If you hold nothing yet, start with the standard your customers name.

Whether it is early, measured rather than guessed

TED is the EU's own procurement portal and its search API is public, so this is checkable rather than asserted. Over the 365 days to 11 August 2026:

Eighteen. Nobody is currently losing public tenders for want of ISO 42001. Certification counts say the same thing from the other side: by the spring of 2026 the published tallies of ISO 42001 certificates worldwide ran to a few hundred organisations, against roughly 96,000 ISO 27001 certificates in the ISO Survey. Accredited certification only became possible in the second half of 2025, when ISO/IEC 42006, the standard that tells certification bodies how to audit against 42001, was published.

What the numbers do not capture is private procurement. Vendor qualification questionnaires from large customers have started to carry an "AI governance" section, and a certificate is the shortest answer to it. That is real demand. It is procurement-driven, it arrives by questionnaire, and it is the reason to do this in 2026 if you do it at all.

What it is not: an EU AI Act route

This is the claim to be careful about, because a lot of marketing makes it.

ISO 42001 does not give presumption of conformity with the EU AI Act. The European standardisation bodies, CEN and Cenelec, wrote a separate European standard for the Act's quality management system: EN 18286.

On 14 October 2026 Commission Implementing Decision (EU) 2026/2274 cited EN 18286:2026 in the Official Journal, the first harmonised standard under the AI Act, and 42001 is not cited. Even that citation is narrower than it sounds. Article 40 presumes conformity with Articles 8 to 15, as far as a cited standard covers them. The quality management system itself is checked against the article that requires it. What EN 18286 does and does not give a provider. The high-risk obligations themselves were deferred by Regulation (EU) 2026/1744 to 2 December 2027 (Annex III systems) and 2 August 2028 (Annex I embedded systems).

What 42001 does give you is the governance scaffolding that any later AI Act work will sit on: roles, an inventory of AI systems, impact assessments, data records, a management review that includes AI. Our own reading maps 42001's controls to each aspect Article 17(1) names, point by point. The defensible sentence is "42001 gets the organisation ready"; the indefensible one is "42001 gets the system compliant".

What it asks for

Clauses 4 to 10 are the Harmonized Structure, the same skeleton as ISO 27001, ISO 9001 and every other current management-system standard: context, leadership, planning, support, operation, performance evaluation, improvement. If you hold ISO 27001 you have built all seven, and the records count for both standards. A scope statement, a policy, objectives, a risk method, competence records, an internal audit, a management review: one set, two certificates.

Three things are new.

The organisation's role for each AI system. 42001 asks you to say, for every system in scope, whether you are its provider, its developer, its deployer or its user, because the obligations differ. A company that deploys a vendor's model and a company that trains its own have different Statements of Applicability, and the scope has to list the systems, not just the business units.

The AI system impact assessment, clause 6.1.4. This is the clause with no ISO 27001 counterpart at all, and it is the reason a 42001 audit is not a 27001 audit with extra controls.

Where 27001 asks what could go wrong for the organisation, 42001 asks what the system does to the people it touches and to society: who is affected, what the intended purpose is, what reasonably foreseeable misuse looks like, and what the consequences of a wrong output are for an individual. An auditor will ask for the assessment of one named system and expect a document, not a filtered risk register.

The 38 controls of Annex A. Grouped A.2 to A.10: AI policy; internal organisation; resources for AI systems (data, tooling, computing, human); impact assessment; the AI system life cycle from objectives through verification, deployment and monitoring; data for AI systems (provenance, quality, preparation); information for interested parties; use of AI systems; and third-party and customer relationships.

As with 27001, you determine which apply and justify the ones that do not in a Statement of Applicability, and Annex B gives implementation guidance for each. The data controls in A.7 are the ones companies underestimate: an auditor will want to see where training and evaluation data came from, how its quality was assessed and how it was prepared, with dates.

What it costs, and why the published range is useless

One search results page for "ISO 42001 certification cost" carried figures from $7,000 to $650,000. Both may be true of somebody. Neither tells you anything about your company. The cost has three parts, and only one of them is set by a rule.

The certification body's audit fee. Under ISO/IEC 42006, the body determines audit time from the number of AI systems in scope, their complexity and the organisation's size, then charges a day rate. Bodies do not publish rates; expect the same order as ISO 27001 auditor days, which in Europe run roughly €1,200 to €2,000 a day.

If the same body already holds your 27001 certificate, ask for an integrated audit: the shared clauses are audited once and the 42001-specific work is added on top, which is days added rather than days doubled.

Your internal time. For a company that already runs an ISMS this is the impact assessments, the AI system inventory, the data records and the 38 control decisions. Weeks of one person's time, not months of a team's. For a company starting from nothing it is building the management system first, which is the ISO 27001 project by another name.

Help, if you buy it. A consultant is selling judgement on scope and on the impact assessments, which is where a first 42001 audit goes wrong. Templates are not the scarce part.

So the honest arithmetic for a 40-person company with an ISO 27001 certificate and two AI systems in scope is a small number of additional auditor days plus a few weeks of internal work. For the same company with no management system, it is the ISO 27001 project plus that.

If you already hold ISO 27001

Do the second standard on the first one's spine. Extend the scope statement to list the AI systems. Add the role determination to the context record. Add "consequences to individuals and to society" to the risk criteria, because 42001 requires it and 27001 does not. Run the impact assessment for each system in scope and keep it as its own document. Decide the 38 controls. Then the internal audit and the management review cover both standards in one cycle, and the surveillance audits align.

The mistake to avoid is a second, separate set of records for one system. Companies that do 27001 with one tool and 42001 with a spreadsheet end up describing the same access control, the same supplier and the same incident twice, and an auditor who sees the two versions disagree has a finding.

Our clause-by-clause coverage for ISO 42001 says, for each clause, what the product holds and where the entries were closed, in the same format as the 27001 register. Read it before you decide anything.

If you hold nothing yet

Start with the standard your customers are asking for. On the numbers above, for almost every company in Europe in 2026 that is ISO 27001. Build it so that the AI standard can be added later: list AI systems in the asset inventory from the beginning, and write the risk criteria with individuals and society as a column even if it stays empty for a year.

Certify 42001 first only if the customer asking for it is the one paying for it. Some are: an AI vendor whose largest customer's procurement requires an AI governance certificate is in exactly that position, and for them 42001 first is rational. Everyone else is buying an option on 2028, and options are cheaper the less you spend on them.

Sources

• TED counts: scripts/ted-by-country.mjs against the TED Search API, 365 days to 11 August 2026; see ISO 27001 or SOC 2 in Europe.
• ISO/IEC 42001:2023, clauses 4 to 10, 6.1.4, Annex A (38 controls, A.2 to A.10), Annex B.
• ISO/IEC 42006:2025, requirements for bodies auditing and certifying AI management systems.
• CEN-CENELEC JTC 21 and EN 18286:2026, cited by Commission Implementing Decision (EU) 2026/2274 of 12 October 2026 (OJ L, 2026/2274, 14.10.2026), read from CELLAR on 17 October 2026; Regulation (EU) 2026/1744 (in force 27 July 2026) deferring Annex III high-risk obligations to 2 December 2027 and Annex I to 2 August 2028. Summaries by Hunton, White & Case and Mayer Brown, July 2026; OJ text to be cited directly before publication.
• Certificate counts: ISO Survey (ISO 27001, ~96,000 certificates); ISO 42001 tallies as of April 2026 from industry counts compiled in our regulatory research. Treat the 42001 figure as an order of magnitude.
• Auditor day rates: How many auditor days does ISO 27001 take.