ISO/IEC 42001 Explained: What That AI Certificate Badge Does and Does Not Prove

A vendor shows you an ISO/IEC 42001 badge and wants a signature by Friday. It is a real achievement and a narrow one: it shows that an organization runs a management system for AI within a stated scope, not that…

FairByDesign The Control File title card: “ISO/IEC 42001 Explained: What That AI Certificate Badge Does and Does Not Prove”

The Control File

You run operations at a 40-person firm, and a vendor wants you to sign by Friday. Their AI tool will screen your job applicants, or draft your client letters, or sort your support tickets. You ask the obvious question: how do we know it’s safe? They point to a badge at the bottom of their trust page. “ISO/IEC 42001 certified.” Your boss reads that as “checked and approved.” If something goes wrong in month three, the person holding that assumption will be you.

Here is our proposition: a 42001 badge is a real promise about how a company runs itself, and almost no promise at all about the answer the AI gives you. The vendor benefits from the gap between those two things. The buyer, and the people the AI decides about, pay for it. A badge you misread is worse than no badge at all.

The claim on the trust page

ISO/IEC 42001 was published in December 2023. ISO describes it as the first international standard for an artificial intelligence management system, and says it sets requirements for “establishing, implementing, maintaining, and continually improving” that system inside an organization that provides or uses AI (International Organization for Standardization [ISO], n.d.-b).

Notice the words management system. Like its better-known cousins for quality (ISO 9001) and information security (ISO/IEC 27001), it is about how an organization runs itself. Has it set policies and objectives for AI? Assigned roles? Assessed risks and impacts? Checked itself through internal audit and management review, and fixed what it found? Its core requirements sit in clauses 4 to 10, and they include completing an AI impact assessment (Sajewski, 2024). It is not a test of a model, a dataset or an answer.

And ISO doesn’t hand out the badge. Its own guidance is blunt: it “does not perform certification or issue certificates” (ISO, n.d.-a). Independent certification bodies do, and the better ones are accredited by a national accreditation body that checks the certifiers. In 2025 ISO published a companion standard, ISO/IEC 42006, setting extra requirements for the bodies that audit and certify against 42001 (ISO, n.d.-c). A certificate typically runs on a three-year cycle, with lighter surveillance audits in between and a full recertification at the end (Sajewski, 2024).

Key idea: certificate scope

Every management-system certificate covers a defined scope: named products, services, sites or activities. Anything outside that scope was not examined, however prominently the badge is displayed. For you, this means the first question about any certificate is not “is it real?” but “does it cover the system that is actually making decisions about me?”

The receipts: two of the biggest names, read line by line

In November 2024, Amazon Web Services announced an accredited ISO/IEC 42001 certification. Read the announcement and the scope is specific: four services, Amazon Bedrock, Amazon Q Business, Amazon Textract and Amazon Transcribe, certified by Schellman Compliance and accredited by the ANSI National Accreditation Board. AWS said the result meant an independent third party “has validated that AWS is taking proactive steps to manage risks” (Duffer et al., 2024). AWS later reported that its first surveillance audit, completed in November 2025, produced no findings (Patil, 2026).

That is a genuine achievement. It is also a statement about a management system around four named services. It says nothing about every AI product AWS sells, or about the application a customer builds on top of them and points at you.

Microsoft’s compliance page for ISO/IEC 42001 lists the services in scope, from Microsoft Copilot and GitHub Copilot to Dragon Copilot and Microsoft Foundry, and points customers to certificates and audit reports on its Service Trust Portal. Its frequently asked questions contain the most honest sentence on the subject we have found in a vendor document. Customers can use Microsoft’s certification in their own compliance work, but they remain “responsible, however, for engaging an assessor to evaluate the controls and processes within your own organization” (Microsoft, 2026).

Translation: the supplier’s certificate covers the supplier. Your deployment, your data, your prompts and your reviewers are yours. Credit to Microsoft for writing that down. It is the sentence most sales decks leave off the slide.

The best case for the badge

Here is the strongest argument against our scepticism, and it is a good one. An accredited 42001 certificate tells you that, within the stated scope, an independent auditor examined the organization’s AI management system against a published set of requirements and found it conformed, at the time of the audit and the surveillance visits since. That means the organization had written AI policies, named responsibilities, a process for assessing AI risks and impacts, internal audits and management reviews, and a way of correcting problems.

For a buyer, that is a meaningful signal of organizational maturity. It is much harder to fake than a values statement. If two vendors are otherwise equal, the certified one has shown its homework to someone who was allowed to say no.

Where the promise runs out

What the badge does not prove is just as specific. It does not prove that any particular output is accurate, fair or lawful. It does not prove that systems outside the scope are governed at all. It does not prove that a control operated on every case, every day; audits sample.

And it does not, on its own, prove compliance with the EU AI Act. Under Article 40 of the Act, the legal presumption of conformity attaches to harmonised standards whose references are published in the EU’s Official Journal (Regulation (EU) 2024/1689, art. 40(1)). The first such reference was published on 14 October 2026: Commission Implementing Decision (EU) 2026/2274 lists one harmonised standard, EN 18286:2026, on quality management systems for AI Act regulatory purposes. Applying it brings a presumption of conformity only for the requirements it covers, and ISO/IEC 42001 is not on the list (European Commission, 2026). With the EU’s main rules for stand-alone high-risk systems now applying from 2 December 2027 under the Digital Omnibus on AI (Hunton Andrews Kurth, 2026; Cuatrecasas, 2026), expect many more 42001 badges before then. Expect some of them to be sold as more than they are.

Chapter 11 of Operational AI Readiness offers a useful ladder. “A human review step is a control” (Bakos, 2026, p. 75). A record of who reviewed which case, with what context, is evidence. Being able to replay a disputed case is reconstruction. Showing that the control worked across the whole period, not just on audit day, is assurance. A management-system certificate mostly tells you the organization has built the first rungs and checks itself. Whether a specific workflow can climb the rest is a question only its own records can answer.

The verdict

For buyers, 42001 is a good filter and a bad finish line. Ask for the certificate itself, the certification body, the accreditation body, the dates and the scope statement, and check whether the service you are buying is named in it. Then do the work no supplier certificate can do for you: classify your own use, decide where human review sits, and make sure your workflow keeps records you could show a reviewer. If you are seeking certification yourself, the honest path is a scope you can actually evidence, not the widest scope that will look good on a slide.

And if you are on the receiving end, a patient whose notes are drafted by an AI scribe or an applicant screened by an AI tool, a certificate on the vendor’s website does not change your rights. It does not mean someone checked the output that concerns you. You can still ask whether AI was involved, ask for a human review, and complain if something is wrong. A badge is a promise about process. It is not an answer about your case.

What we still don’t know

We have not seen the full AWS or Microsoft certificates, scope statements or audit reports, which sit behind customer portals; we rely on what the companies have published about them. The statements about surveillance audits with no findings are the companies’ own. We have not read the paid text of ISO/IEC 42001 or 42006 and describe them only from public summaries. Nothing here suggests that either company’s certification is anything other than what it says.

The biggest gap is the one that matters most: nothing we found shows whether certified organizations have fewer AI incidents than uncertified ones. That evidence does not yet exist in the public record. So here is the question we will keep asking: when a certified system gets a case wrong, will anyone be able to tell?

Questions you can ask this week

  • If an AI system affects you: Was AI involved in this decision, and can a person review it? A certificate does not replace that answer.
  • If an AI system affects you: Which organization is responsible for this system as I experienced it: the vendor, or the company that used it on me?
  • If you buy AI: Is the exact service we use named in the certificate’s scope, who issued it, who accredited them, and when does it expire?
  • If you run AI: If a reviewer picked one disputed case from last month, could we show what our system did and who checked it, from records rather than memory?

Go further. Operational AI Readiness is FairByDesign’s free field manual for AI that is already live. Chapter 11 sets out what auditors, regulators and internal reviewers actually ask to see, and why “legibility under pressure is the real standard” (Bakos, 2026, p. 75). Read it free, or test one workflow with the free readiness assessment.


Support independent explanation. Badges, certificates and trust pages are written by the people being assessed. FairByDesign reads the scope line so you do not have to, and we can only do that if we stay independent of the companies we explain. Give once ($5, $10, $25, $50 or any amount), give monthly ($10 or $25), or become a patron: fairbydesign.org/patrons-2. Support never buys editorial control.

References

Bakos, V. (2026). Operational AI readiness: A practitioner guide to governing AI that is already live. FairByDesign. https://fairbydesign.org/operational-ai-readiness/

Cuatrecasas. (2026, July). Digital Omnibus on AI: Key changes to the Artificial Intelligence Act [Legal Flash]. https://www.cuatrecasas.com/en/spain/intellectual-property/art/key-aspects-ai-omnibus-regulation

Duffer, S., Singh, A., & Hallinan, P. (2024, November 25). AWS achieves ISO/IEC 42001:2023 Artificial Intelligence Management System accredited certification. AWS Machine Learning Blog. https://aws.amazon.com/blogs/machine-learning/aws-achieves-iso-iec-420012023-artificial-intelligence-management-system-accredited-certification

European Commission. (2026). Commission Implementing Decision (EU) 2026/2274 of 12 October 2026 [harmonised standards for high-risk artificial intelligence systems drafted in support of Regulation (EU) 2024/1689]. Official Journal of the European Union, L, 2026/2274, 14 October 2026. http://data.europa.eu/eli/dec_impl/2026/2274/oj

Hunton Andrews Kurth. (2026, July 28). EU Digital Omnibus on AI enters into force. Privacy & Cybersecurity Law Blog. https://www.hunton.com/privacy-and-cybersecurity-law-blog/eu-digital-omnibus-on-ai-enters-into-force

International Organization for Standardization. (n.d.-a). Certification. Retrieved October 15, 2026, from https://www.iso.org/certification.html

International Organization for Standardization. (n.d.-b). ISO/IEC 42001:2023 Information technology — Artificial intelligence — Management system. Retrieved October 15, 2026, from https://www.iso.org/standard/42001

International Organization for Standardization. (n.d.-c). ISO/IEC 42006:2025 Information technology — Artificial intelligence — Requirements for bodies providing audit and certification of artificial intelligence management systems. Retrieved October 15, 2026, from https://www.iso.org/standard/42006

Microsoft. (2026). ISO/IEC 42001:2023 Artificial Intelligence Management System standards. Microsoft Learn. https://learn.microsoft.com/en-us/compliance/regulatory/offering-iso-42001

Patil, A. (2026, February 26). AWS successfully completed its first surveillance audit for ISO 42001:2023 with no findings. AWS Security Blog. https://aws.amazon.com/blogs/security/aws-successfully-completed-its-first-surveillance-audit-for-iso-420012023-with-no-findings

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). (2024). Official Journal of the European Union, L, 2024/1689. http://data.europa.eu/eli/reg/2024/1689/oj

Sajewski, M. (2024, November 25). What are the ISO 42001 requirements? Cloud Security Alliance (originally published by Schellman). https://cloudsecurityalliance.org/articles/what-are-the-iso-42001-requirements

Leave a Reply