Executive summary
Dec 2023
ISO/IEC 42001 published, the first certifiable international standard for AI management systems
~350
organisations certified worldwide by spring 2026, by one industry count; there is no official register
17 Feb 2025
the Singapore Accreditation Council opened its ISO/IEC 42001 accreditation programme
2 Dec 2027
when EU AI Act duties for stand-alone high-risk systems now apply, after the 2026 Digital Omnibus
Key conclusions
- A 42001 certificate covers an organisation's AI management system within a stated scope. It does not certify that any model is accurate, fair or safe.
- Each new deployment strengthens the case for it. Organisations run more systems than anyone can review by hand, AI arrives inside bought software, and systems drift after handover.
- The certification infrastructure is now in place. ISO/IEC 42006 sets the rules for certifiers, and Singapore has accredited certification bodies since February 2025.
- 42001 does not satisfy the EU AI Act on its own. It does provide the base for the Act's quality-management requirement, and the 2026 delay gives organisations time to build that base now.
AI has moved from the pilot budget into the operating budget. Models now sit inside customer service, credit decisions, scheduling, inspection and maintenance planning, and most of them reached production before anyone had built the governance around them. Buyers, regulators and boards now take it as given that an organisation uses AI. What they ask is whether anyone can show how that AI is controlled.
ISO/IEC 42001:2023, published in December 2023, is the first international standard that answers that question in a form an auditor can test [1]. It specifies requirements for an AI management system: the policy, roles, risk and impact assessments, controls and review cycles an organisation runs around the AI it builds or uses. Certification means an accredited third party has examined that system and found it in operation.
A 42001 certificate covers the management system and does not certify any model
The certificate is issued to an organisation's management system. It is not issued to a model. It does not say a system's outputs are accurate, unbiased or safe. It says the organisation has identified what could go wrong and assessed the effect on the people its AI affects. It has assigned an owner to each risk and put controls in place. It checks that the controls work, corrects them when they fail, and can produce evidence of each step.
The distinction matters because marketing material often blurs it. There is no such thing as an ISO 42001 certified model. Every certificate carries a scope statement naming the organisation, the sites and the AI systems it covers. Read the scope before you rely on the logo.
Four pressures make certification more valuable with each new AI deployment
Each of these pressures grows with every additional deployment:
- The number of systems outgrows manual review. One model can be reviewed by hand, but forty spread across five departments cannot. A management system keeps track of what is running, who owns each system and when each was last checked, at the scale the organisation actually operates.
- AI arrives through suppliers. Much of the AI an industrial firm runs is embedded in software it bought: ERP and WMS modules, vision systems, copilots. Annex A of the standard includes controls on third-party and customer relationships, so a certified buyer asks better questions, and a certified supplier has the answers ready.
- Systems drift after handover. A one-time sign-off proves a system worked once. A certified management system has to show monitoring, internal audit, management review and corrective action, and the certification body comes back for surveillance audits through the certificate's three-year cycle to check that they are still happening.
- Customers need evidence they can rely on. A customer cannot audit every supplier's AI. Accredited certification gives procurement teams, insurers and regulators one independent examination to rely on, instead of each sending its own questionnaire.
The certification infrastructure is now in place
For its first year and a half, 42001 had a credibility gap of its own: the rules for the bodies certifying against it were still in draft. ISO/IEC 42006:2025, published in July 2025, closed that gap. It adds AI-specific requirements to ISO/IEC 17021-1 for bodies that audit and certify AI management systems, and accreditation bodies can use it as the criteria for accrediting those certifiers [2].
Singapore moved early. The Singapore Accreditation Council opened its ISO/IEC 42001 accreditation programme on 17 February 2025, assessing certification bodies against ISO/IEC 17021-1 and the then-draft 42006 [3]. Within days, SGS announced an SAC-accredited 42001 certificate for Changi Airport Group, scoped to five named AI applications ranging from passenger transfer estimation to generative AI search [4]. Other organisations should copy that approach, because the scope names the systems as well as the company.
Among AI developers, Anthropic announced an accredited 42001 certificate in January 2025, issued by Schellman, a certification body accredited by ANAB [5]. A buyer evaluating a model provider can now ask for a certificate and a scope instead of accepting a policy page.
A vendor's 42001 claim may mean certified, aligned or only in progress
The wording on a vendor's slide often claims more than the evidence supports. Here is what each claim means and what to ask for:
| What the vendor says | What it means | What to ask for |
|---|---|---|
| ISO/IEC 42001 certified | An accredited body audited the management system and issued a certificate within a stated scope. | The certificate, its scope statement, the issuing body, that body's accreditation, and the expiry date. |
| Aligned with, or built to, ISO/IEC 42001 | The organisation used the standard as a reference. No third party has checked it. | The Statement of Applicability and the latest internal audit report. |
| Our model is ISO 42001 certified | Not possible as stated. The standard certifies a management system, not a model. | Which organisation holds the certificate, and whether the product you are buying is inside its scope. |
| Certification in progress | Anything from a gap assessment to a booked Stage 2 audit. | The certification body's name and the Stage 1 and Stage 2 audit dates. |
Few organisations are certified so far
There is no central register. ISO does not certify organisations, and each certification body keeps its own directory [6]. Pieced together from certification-body announcements and company press releases, one industry count put certified organisations worldwide at around 350 by spring 2026 [6]. Treat that as an order of magnitude rather than a measurement.
Even at several times that figure, it is small next to the number of organisations now running AI in production. That is why the certificate still sets an organisation apart. The advantage will narrow as procurement teams start to require certification instead of merely welcoming it.
42001 supports EU AI Act compliance but does not satisfy the Act alone
The EU AI Act is the regulation most likely to reach a Singapore manufacturer or service provider, usually through its European customers. The Digital Omnibus, Regulation (EU) 2026/1744, entered into force on 27 July 2026 and moved the high-risk deadlines. Stand-alone high-risk systems listed in Annex III are now covered from 2 December 2027. AI in products already governed by EU product legislation (Annex I) follows on 2 August 2028 [7][8]. The main reason given was that the harmonised standards the Act relies on were running late [7].
The delay is no reason to wait. However, 42001 cannot replace compliance with the Act. It is not a harmonised standard under the Act, so it gives no presumption of conformity. The Cloud Security Alliance's April 2026 analysis puts it plainly: 42001 alone does not satisfy the Act, and a provider relying only on it carries the full burden of proof if challenged. The same analysis lists the gaps, including serious-incident reporting timelines and change management for systems that keep learning [9].
What 42001 does provide is the organisational base that the Act's quality-management requirement in Article 17 builds on: a risk process, documented roles, supplier controls, monitoring and corrective action. The draft European quality-management standard, prEN 18286, maps its own requirements back to 42001's Annex A controls, so a certified organisation can reuse what it has built [9]. An organisation that starts now can have a certified, operating management system in place when the Annex III obligations apply. One that waits for the harmonised standards will be starting from nothing against the same deadline.
Key dates from the standard's publication to the EU deadlines
The key dates, in order. The period between the arrival of the certification infrastructure and the EU deadlines is the time available to prepare.
Dec 2023
ISO/IEC 42001 published [1]
13 Jan 2025
Anthropic announces an accredited 42001 certificate [5]
17 Feb 2025
Singapore Accreditation Council opens its 42001 accreditation programme [3]
Feb 2025
SGS certifies Changi Airport Group, scoped to five named AI applications [4]
Jul 2025
ISO/IEC 42006 sets the rules for certification bodies [2]
27 Jul 2026
Digital Omnibus, Regulation (EU) 2026/1744, enters into force [8]
2 Dec 2027
EU AI Act duties apply to stand-alone high-risk systems (Annex III) [7]
2 Aug 2028
EU AI Act duties apply to AI in regulated products (Annex I) [7]
Four first steps apply whether or not you seek certification
For an operator with AI already in production, the first steps are the same whether or not certification is the goal:
- Inventory every AI system in use, including the ones inside software you bought, and name an owner for each.
- Run an AI system impact assessment (clause 6.1.4) on the systems that touch customers, safety, money or people's jobs. Teams that already work with ISO/IEC 27001 usually find their largest gap here, because 27001 has no equivalent. It asks about risk to the organisation and does not cover the consequences for the people a system affects.
- Set the scope: which systems and which sites the certificate will cover. Choose a scope that the organisation actually operates. A narrow scope in real use is worth more than a broad one that exists only on paper.
- Write the Statement of Applicability against Annex A, then audit it internally before a certification body does.
Certification shows that an organisation knows and controls the AI it runs
As AI spreads, it becomes more costly not to know which systems you run. ISO/IEC 42001 certification is currently the only way to show, through an accredited third party, that an organisation knows what AI it runs, who owns it, what could go wrong and what happens when it does. It is not a quality mark for a model. Anyone who presents it as one weakens the credibility the standard is meant to build.
Our free ISO 42001 readiness check maps seventeen questions to clauses 4 to 10 and Annex A and returns a gap list ordered by audit exposure. Our assurance practice takes an organisation from that gap list through to its Stage 1 and Stage 2 audits.
