The EU AI Act Has Its First Harmonised Standard. EN 18286 Supports Article 17. ISO 42001 Alone Gives No Presumption.

1. What happened on October 14, 2026?

On October 14, 2026, the EU AI Act got its first harmonised standard. Commission Implementing Decision (EU) 2026/2274, adopted on October 12, was published in the Official Journal that day and entered into force on publication. Its annex lists exactly one reference: EN 18286:2026, “Artificial intelligence – Quality management system for EU AI Act regulatory purposes.”

The standard was drafted by CEN and CENELEC, the two European standardisation organizations, through their joint technical committee on AI, CEN/CLC/JTC 21. According to the Commission’s Interoperable Europe portal, CEN and CENELEC adopted it on July 22, 2026. Until the citation it was a voluntary European standard without the Article 40 presumption. The Official Journal citation is what gives it that legal effect, and CEN and CENELEC’s own announcement calls it the first harmonised standard in support of the AI Act.

Here is the definition in one paragraph. EN 18286:2026 is the European harmonised standard for the quality management system that providers of high risk AI systems must run under Article 17 of the EU AI Act. Since its reference was published in the Official Journal on October 14, 2026, conforming to it gives a presumption of conformity with the requirements it covers.

For providers of high risk AI, this is the first time a published standard offers a legal shortcut to showing compliance with any part of the Act. It is one shortcut, for the quality management requirements the standard covers. Everything below is about using it well without mistaking it for more than it is.

2. What does “harmonised standard” and “presumption of conformity” mean?

A harmonised standard is a European standard that the Commission has asked for, assessed against the law, and cited in the Official Journal; conforming to it creates a legal presumption that you meet the requirements it covers. Article 40(1) of Regulation (EU) 2024/1689, the AI Act, sets that rule for high risk AI systems, “to the extent that those standards cover those requirements.”

Recital 6 of Decision 2026/2274 adds the timing: the presumption runs “from the date of publication of the reference” in the Official Journal. That is why the Decision entered into force the day it was published. For EN 18286, the presumption starts on October 14, 2026, not in July when CEN and CENELEC made the text available.

Two limits matter. First, standards are voluntary. A provider can meet Article 17 another way; it then has to show on its own how it complies. Article 17(1)(e) itself expects providers to record which standards they apply and, where harmonised standards are not applied in full, “the means to be used” to comply. Second, EN 18286 is not an AI Act certificate.

The presumption reaches only what the standard covers. A quality system addresses processes connected to risk management, data, testing and monitoring, but running those processes well is not the same as showing independent conformity with every substantive requirement in Articles 9 to 15, such as accuracy, robustness or human oversight design. Each of those has its own requested standards area, and none has a cited standard yet. The precise legal coverage of EN 18286 has to be checked in its Annex Z and against the Official Journal citation.

3. What does EN 18286 cover?

EN 18286 covers the quality management system that Article 17 requires every provider of a high risk AI system to put in place. The Decision places it under point 9 of the Commission’s standardisation request, the area covering quality management systems for providers of AI systems, including post market monitoring processes.

Article 17(1) requires a system “documented in a systematic and orderly manner in the form of written policies, procedures and instructions,” and lists 13 aspects, points (a) to (m). In plain terms they fall into five groups.

• Compliance strategy: a strategy for regulatory compliance, including conformity assessment and how modifications to the system are managed.
• Building the system: design, design control and verification; development, quality control and quality assurance; and testing and validation before, during and after development, with defined frequency.
• Data and risk: data management from acquisition and labelling to retention, and the Article 9 risk management system.
• After release: post market monitoring under Article 72, serious incident reporting under Article 73, and communication with authorities, notified bodies, customers and others.
• Running the system: record keeping, resource management including security of supply, and an accountability framework that names who is responsible for each aspect.

Two of those points reach across the rest of the Act. Point (g) pulls the Article 9 risk management system into the quality system, and point (h) pulls in post market monitoring. A quality system that conforms to EN 18286 is therefore the container in which much of your other compliance work lives. The public announcements identify Article 17 as its focus; providers should use Annex Z to establish the exact requirements, and parts of requirements, it supports.

The Commission’s standardisation request requires each harmonised standard to indicate, in an Annex Z, the relationship between its clauses and the AI Act requirements, and to identify which requirements it aims to cover and which it does not. That annex is part of the standard itself, which you obtain from a national standards body. We have not reproduced it here, and nobody should claim clause by clause coverage without reading it.

One connection deserves attention. The request says there is no separate standard for Article 11 technical documentation and that CEN and CENELEC must cover Article 11 in other standards, “particularly on the quality management system.” That does not establish that EN 18286 gives a presumption for Article 11. It does mean you should check Annex Z on that point rather than assume either way.

4. Who needs EN 18286 and who does not?

EN 18286 matters to providers of high risk AI systems, not to every company that uses AI. Article 17 is a provider obligation. A deployer is not subject to Article 17 merely because it uses a high risk system, although it has its own duties under the Act. It should separately assess whether putting its name or trademark on the system, making a substantial modification, or changing a system’s intended purpose makes it a provider under Article 25(1).

A system is high risk in one of two ways under Article 6. It falls under Article 6(2) and Annex III, which covers uses such as employment, education, credit and access to essential services, subject to the Article 6 qualifications, including the Article 6(3) derogation for a system that does not pose a significant risk of harm, for example because it performs only a narrow procedural task. Or, under Article 6(1), it is a safety component of a product, or is itself a product, covered by the EU product laws in Annex I, and that product must undergo a third party conformity assessment. Annex I includes the Medical Devices Regulation, (EU) 2017/745.

Integration into an Annex I product is not enough on its own; the Article 6(1) conditions must be met. Annex I is also split into two sections. For products under Section B, Article 2(2) applies only Article 6(1), Article 60a and Articles 102 to 112 of the AI Act, so those sectors are treated differently and need their own analysis.

Location does not take a US company out of scope. Article 2(1)(a) applies the Act to providers placing AI systems on the EU market or putting them into service in the EU “irrespective of whether those providers are established or located within the Union or in a third country.” Article 2(1)(c) also reaches providers in a third country where the system’s output is used in the EU.

So the first question is classification, not standards. A SaaS company selling résumé screening to European employers is likely an Annex III provider. A US contractor building a hiring tool only for American federal agencies is likely not in scope for this purpose. A digital health company whose software is a medical device requiring notified body review is likely an Annex I provider. Each needs a documented answer before EN 18286 becomes a priority.

5. How does EN 18286 relate to ISO/IEC 42001?

ISO/IEC 42001 is not a harmonised standard under the AI Act, so certifying to it does not by itself give a presumption of conformity with any AI Act requirement; EN 18286 does, for the parts it covers. That is the single most important point for organizations that have invested in ISO/IEC 42001, and it is easy to blur.

ISO/IEC 42001 is an international, certifiable management system standard for AI that applies across an organization’s AI activities. EN 18286 is a European standard written, in its own title, “for EU AI Act regulatory purposes,” and the Commission cited it against a provider obligation that attaches to each high risk AI system. ISO/IEC 42001 provides an organization level AI management framework. EN 18286 is directed at the provider’s regulatory quality system for its high risk AI systems. One integrated management system may support multiple products, provided its scope and controls cover the applicable legal requirements.

For federal contractors, the distinction has a second dimension. ISO/IEC 42001 can be useful supporting evidence in US procurement, though it does not by itself satisfy the GSA AI clause we covered in our federal contractor analysis. We are not aware of any recognized role for EN 18286 in US federal procurement. In our view, a firm selling into both markets should run one management system and layer the EU specific product controls on top, not run two programs.

6. What does Article 17 now say about small companies?

Article 17(2) has always required the quality system to be proportionate to the size of the provider’s organization; the Digital Omnibus now names SMEs, start ups and small mid caps explicitly. Regulation (EU) 2026/1744, adopted July 8, 2026, published July 24 and in force July 27, replaced the paragraph with this text.

The new text says implementation of the quality system aspects shall be proportionate to the size of the provider’s organization, in particular where the provider is an SME, including a start up, or an SMC. The next sentence keeps the floor: providers “shall, in any event, respect the degree of rigour and the level of protection required.” Proportionality is not new. The explicit SME and SMC wording is the Omnibus change.

The definitions are concrete. An SME, under Recommendation 2003/361/EC, employs fewer than 250 people and has turnover of no more than €50 million or a balance sheet of no more than €43 million. A small mid cap, under Recommendation (EU) 2025/1099, is not an SME, employs fewer than 750 people, and has turnover of no more than €150 million or a balance sheet of no more than €129 million. Headcount and financial figures are calculated under each recommendation’s rules, including aggregation for partner and linked enterprises, so size is not judged on your own payroll and revenue alone.

The Omnibus also widened a second relief. Article 63(1) now lets SMEs, including start ups, comply with “certain elements” of the Article 17 quality system “in a simplified manner,” provided they have no partner or linked enterprises within the meaning of Recommendation 2003/361/EC. Before the Omnibus that option was limited to microenterprises. The Commission must develop guidelines on which elements may be simplified “without affecting the level of protection.” Check for those guidelines before you scope a simplified system; we are not aware of their publication as of October 22, 2026.

Proportionality changes the size of the system, not its existence. A 40 person provider can run a lean quality system with fewer documents and combined roles. It cannot skip post market monitoring or serious incident procedures because it is small.

7. What if you already run a sector quality system, such as for medical devices?

Providers already subject to a quality management obligation under sector specific EU law may build the Article 17 aspects into that existing system. Article 17(3) says so directly. It is the most practical sentence in the Article for healthcare technology companies.

A medical device manufacturer already runs a quality system under the Medical Devices Regulation. If its AI software is a high risk system because it is a device, or a safety component of one, requiring third party conformity assessment, the AI elements can be added to that system rather than built as a second one. The high risk obligations for those Annex I products apply from August 2, 2028.

Article 17(4) gives a similar route to financial institutions already subject to EU internal governance rules. For them, the quality system obligation is deemed met by complying with those rules, except for points (g), (h) and (i): risk management, post market monitoring and serious incident reporting. Those three still have to be addressed, and the paragraph says harmonised standards “shall be taken into account.”

US healthcare readers should keep the patient side separate. The AI Act quality system governs the provider’s product. It does not replace HIPAA, FDA requirements or state rules on clinical AI, which we covered in our patient intake chatbot analysis.

8. Where do the other AI Act standards stand?

As of October 22, 2026, the Commission’s AI harmonised standards page lists EN 18286 as the cited standard, and no cited references appear there for the other nine requested areas, so none of them gives a presumption of conformity yet. The Commission first asked CEN and CENELEC for AI standards in 2023. On June 23, 2025 it replaced that request with Implementing Decision C(2025) 3871, aligned with the final text of the Act and covering ten areas, all for high risk AI systems.

“None yet” means no Official Journal citation, not that no technical work exists. Each area may produce one or more harmonised standards or other deliverables, and some may be finished but not yet cited; check individual stages with CEN and CENELEC. The request’s own annex says the ten areas must not be “isolated from each other” but form part of one system, which is why the quality standard is a sensible place to start. The Commission’s harmonised standards page for AI is where the next citations will be listed. We are not predicting when.

The Omnibus added a further instruction. Article 40(2) now tells the Commission to request standards that “facilitate the joint compliance and presumption of conformity” with the AI Act and the EU product laws listed in Annex I. For device makers and machinery builders, that points toward standards serving both rulebooks at once.

General purpose AI models sit outside this picture. The 2025 request covers high risk AI systems only. Under Article 53(4), general purpose model providers may rely on codes of practice “until a harmonised standard is published,” for example the GPAI Code of Practice we discussed in our analysis of downstream GPAI obligations. A code is not the only route: providers that do not adhere to an approved code or comply with a harmonised standard must demonstrate alternative adequate means of compliance for assessment by the Commission.

9. Does EN 18286 change the AI Act deadlines?

No. The deadlines come from the Digital Omnibus, not from the standard, and the citation of EN 18286 did not move them. Article 113, as amended, applies Sections 1, 2 and 3 of Chapter III, which include Article 17, except Article 6(5), from December 2, 2027 for systems classified as high risk under Article 6(2) and Annex III, and from August 2, 2028 for systems classified under Article 6(1) and Annex I.

Existing systems are subject to a transitional rule. Under Article 111(2), high risk systems placed on the market or put into service before the Chapter III application date are covered only if, from that date, they are subject to significant changes in their designs, while providers and deployers of high risk systems intended for use by public authorities must comply by August 2, 2030. The application date alone is not a complete assessment of an individual system’s obligations.

The Omnibus explains why the dates moved. Its recitals point to “the delayed availability of standards, common specifications, and alternative guidance” and the delayed setup of national authorities. The postponement was granted because standards were late. It was not a promise that every standard would be ready before the new dates.

That leaves a planning gap. On today’s facts, Annex III providers have about 13 months and one cited standard. If the risk management or logging standards are not cited in time, providers must comply with Articles 9 and 12 directly from the Act’s text and document how. Article 17(1)(e) anticipates exactly that case.

The transparency obligations under Article 50 run on their own clock, which we covered in our Article 50 guide for SaaS companies. Nothing in Decision 2026/2274 affects them.

10. What should a small AI provider do in the next 90 days?

A small provider should spend the next 90 days answering the classification question, sizing a quality system, and building on the Act’s text rather than waiting for more standards. Five steps cover it.

1. Decide whether you are a high risk provider

List every AI system you place on the EU market or whose output is used in the EU. For each one, record whether it falls under Annex III or is a safety component of an Annex I product requiring third party assessment, and whether you are the provider or a deployer. EN 18286 matters to providers of high risk systems, not to every AI user. Write the reasoning down, because the classification is the first thing a customer or regulator will ask about.

2. If you run ISO/IEC 42001, map it against Article 17

Walk the 13 points in Article 17(1) and mark what your 42001 system already produces as evidence and what it does not. Product specific items, such as design verification for a particular system, modification management and serious incident procedures, are where gaps usually sit. Then plan the gap to EN 18286, using its Annex Z for legal coverage and a separate gap assessment for your 42001 system. Do not assume 42001 covers it.

3. Size the system honestly and reuse what you have

If you are an SME or small mid cap, document that status and scale the quality system to it under Article 17(2), without lowering the level of protection. Watch for the Commission’s Article 63 guidelines if you have no partner or linked enterprises. If you already run a sector quality system, such as under the Medical Devices Regulation, build the AI elements into it under Article 17(3).

4. Do not wait for the other nine standards

The obligations apply from December 2, 2027 for Annex III systems and August 2, 2028 for Annex I systems, subject to the transitional rule for existing systems, whether or not every standard is ready. Build risk management, data governance, logging and human oversight against Articles 9 to 15 now, and record under Article 17(1)(e) the means you use where no harmonised standard applies.

5. Get the standard and track the Official Journal

Obtain EN 18286 through your national standards body before you build to it, so you are working from the clauses and the requirement mapping rather than summaries. Assign someone to watch the Commission’s harmonised standards page for the next citations. The NIST AI RMF remains a sound structure for the risk work underneath, but only a cited harmonised standard carries the EU presumption.

This article is provided for general informational purposes and reflects the state of the law as of October 2026. It is not legal advice. Regulations in this area are changing quickly. Confirm current requirements and consult qualified counsel before making decisions for your organization.