ISO/IEC 42001 certification and EU AI Act Article 17: what each one covers
Article 17 of the EU AI Act requires providers of high-risk AI systems to operate a quality management system. This piece is for those providers and their quality and compliance leads: what an ISO/IEC 42001 certificate contributes, where its reach ends, and who the obligation applies to.
By Reinhardt Mühlhäusser · Published
- Written against
- Regulation (EU) 2024/1689 · ISO/IEC 42001:2023 · ISO/IEC 42005:2025 · ISO/IEC 42006:2025
- Last updated
If you hold an ISO/IEC 42001 certificate, or plan to, a fair question follows: what does it contribute toward Article 17? The answer matters to two groups. Providers of high-risk systems need to know what is still left to build. Organizations that use AI without providing a high-risk system need to know whether the obligation reaches them at all. The text below sets out what Article 17 covers, how proportionality and the amended dates work, what certification adds, and where it stops. It closes with two companion standards: ISO/IEC 42005 for impact assessment and ISO/IEC 42006 for the bodies that certify an AI management system.
A certified management system is not a conformity assessment
Permalink to “A certified management system is not a conformity assessment”Article 17 of the EU AI Act requires providers of high-risk AI systems to operate a quality management system, documented as written policies, procedures and instructions, and covering at least: a strategy for regulatory compliance including conformity assessment and the management of modifications; design and design-control techniques; development, quality control and quality assurance; examination, test and validation procedures; the technical specifications and standards to be applied; data management; the risk management system the Regulation requires; post-market monitoring; incident reporting; communication with competent authorities, notified bodies, other operators and customers; record-keeping; resource management; and an accountability framework.
Implementation is proportionate to the size of the organization — but proportionality reduces the apparatus, not the protection: the degree of rigour and the level of protection required must be met in any event. Providers already subject to quality-management-system obligations under sectoral Union law may incorporate these aspects into the system they already operate rather than running a second one alongside it. These obligations apply from 2 December 2027 for Annex III systems and from 2 August 2028 for systems embedded in products under Annex I, as amended by Regulation (EU) 2026/1744. What was not moved is the conformity machinery: the notified-body regime, the standards process, conformity assessment and registration all continue on their original dates.
Neither standard delivers this obligation on its own.
Quality management supplies the design-control, testing and release spine and nothing AI-specific. An AI management system supplies the risk, data and monitoring content and not the per-product quality assurance or the conformity mechanics. Certification to ISO/IEC 42001 is real assurance about how an organization governs AI, and it is genuine evidence toward several of the aspects above. What it does not do is establish that a product conforms, and it cannot discharge the obligation. "Certified, therefore ready" is an overclaim.
Certification is worth having, and it is not the point
Permalink to “Certification is worth having, and it is not the point”Two things are true at once, and the case for certification only holds when both are stated.
Certification is worth having. An external audit checks whether the management system actually does what its documents say, and the answer is not always yes. Preparing for the audit is worth most of the fee on its own.
But no regulation requires it, and it does not make AI pay off. A certified organization can still run AI that returns nothing, because the certificate says how decisions are made — not whether they were good decisions, and not what came of them. It answers one question, about governance, and answers it well. Buy it for that, and expect nothing else from it.
And the obligation reaches providers of high-risk systems — not every organization using AI.
A company using AI inside its own development work, in engineering assistance or generative design tooling, does not become a provider of a high-risk AI system by doing so, and this requirement does not reach it. Other duties still do, including the EU AI Act's obligation on providers and deployers alike to take measures supporting the AI literacy of the people operating these systems. And the quality case for that company is real and entirely different: AI-generated material entering a technical file or a design record is an engineering-integrity problem before it is a documentation one — a fabricated test result or a wrong tolerance threatens the conformity assessment its actual product depends on. That is worth controlling. It is not this obligation, and treating it as this obligation is how organizations buy remediation for a duty they never had.
- 01
ISO/IEC 42005 — impact assessment
Published in 2025. The method for assessing what an AI system does to the people and groups it affects — the input that makes a risk register about consequences rather than components.
- 02
ISO/IEC 42006 — certification bodies
Published in 2025. Requirements for the bodies that audit and certify an AIMS, building on ISO/IEC 17021-1. It matters because it determines what an auditor will actually look for.
Worth checking first is which side of the line you're on. If you provide a high-risk AI system, a useful next step is to map each Article 17 aspect against what your management systems already cover. For Annex I Section A products, Article 2(13) allows the Article 17 duty to be limited where sectoral law protects at least as well, in cases set by delegated act. An ISO/IEC 42001 certificate provides evidence toward several of those aspects. It cannot discharge the obligation, and the per-product quality assurance and the conformity mechanics still need to be covered separately. If you use AI in your own development work without providing a high-risk system, the quality question is different, and still worth controlling. Either way, the approach is the same: one integrated management system.
Get direction and delivery working together.
Tell us what governance exists today and where AI actually runs. We will respond with a read on how the direction and the management system line up, and a sensible first step — often an ISO/IEC 42001 gap analysis.