For the first time, the AI Act includes a harmonized standard. Anyone who implements it is presumed to be in compliance under Article 17. And anyone who implements it but cannot demonstrate compliance receives no such presumption.
On October 19, 2026, the European Commission announced the publication in the Official Journal of the European Union of the reference to EN 18286:2026, Artificial Intelligence. Quality management system for EU AI Act regulatory purposes, developed by CEN and CENELEC. It is the first harmonized standard supporting Regulation (EU) 2024/1689, and the Commission has already indicated that others will follow.
It does not change any dates or add any requirements. What does change is how one of the core requirements of high-risk systems—the quality management system—is demonstrated.
What exactly has been published?
Implementing Decision (EU) 2026/2274
Commission Decision of October 12, 2026, on harmonized standards for high-risk AI systems developed in support of the AI Act.
EN 18286:2026
Quality management system for regulatory purposes under the AI Act. Developed by CEN and CENELEC at the request of the Commission.
Presumption of Conformity
This regulation covers the requirements of Article 17. It applies only to providers of high-risk AI systems. It does not affect general-purpose models.
What is a harmonized standard, and what does it provide a presumption of conformity for?
The AI Act sets out requirements but does not specify how to meet them. That is the role of harmonized standards: the Commission mandates a technical standard from the European standardization bodies, publishes its reference in the Official Journal, and, from that point on, Article 40 of the Regulation presumes that anyone who complies with the standard also complies with the legal requirements that the standard covers.
| Appearance | No harmonized standard | In accordance with EN 18286 |
|---|---|---|
| Interpretation | The supplier interprets Article 17 on its own and defends its interpretation before the authority or the notified body. | The supplier implements the standard, and the burden of proving that the QMS does not comply falls on the party challenging it. |
| Audit | Every evaluation starts from scratch: what should have been there, what is there. | It is compared to a standard text, clause by clause. |
| Systems | The AI SGC coexists with the quality, sector-specific, and safety SGCs as separate systems. | The standard is designed to build upon existing standards and integrate them. |
The presumption is neither automatic nor absolute. It meets the requirements of Article 17 set forth in the regulation, but only if implementation is genuine and demonstrable. That is where most organizations run into trouble.
What a quality management system must include
Article 17 requires providers of high-risk systems to have a QMS that is documented in a systematic and orderly manner through written policies, procedures, and instructions. Among other things, it must cover:
| Scope | What Article 17 Requires | What an auditor asks |
|---|---|---|
| Compliance | A regulatory compliance strategy, including procedures for conformity assessment and change management. | Which version of the policy is currently in effect, and who approved it? |
| Design and Development | Techniques, procedures, and systematic actions for the design, control, and verification of the system. | What has changed between the version that was evaluated and the one currently in production? |
| Data | Data management systems and procedures: collection, labeling, storage, and retention. | What specific data was used to train the model currently in use? |
| Risks | The risk management system referred to in Article 9. | When was it last reviewed, and what decisions came out of that review? |
| Post-Marketing | The post-marketing surveillance system under Article 72 and the reporting of serious incidents under Article 73. | When was the incident detected, who reviewed it, and when was it reported? |
| Records | Document and Information Management, and Records Retention. | Can you prove that this record has not been altered since then? |
| Liability | An accountability framework outlining the responsibilities of management and the rest of the staff. | Who was in charge of this system on the day of the incident? |
EN 18286 translates that list into an operating system. And, according to the Commission itself, it does so based on the AI system’s life cycle rather than on the organization—which is how traditional management standards are structured.
It does not replace your ISO. It builds upon it.
The Commission describes it as the only AI QMS standard with a presumption of conformity, designed to complement ISO/IEC 42001, ISO 9001, or a sector-specific QMS such as ISO 13485, not to replace them.
For an organization certified to ISO/IEC 42001, EN 18286 is the layer that connects its AI management system to the legal requirement. For an organization certified to ISO 9001 but not to 42001, it is the way to expand its existing quality management system (QMS) to cover Article 17. And for regulated sectors, the standard expressly allows for the integration of management systems rather than maintaining multiple systems in parallel.
The practical implication for CIOs, CDOs, and the Legal Department is that the work done on ISO standards is not lost. What is needed is the ability to demonstrate: which version of each procedure is currently in effect, who approved it, what records exist, and that none of them have been altered since then.
A presumption is only as good as the evidence
A QMS lives in documents: policies, procedures, review minutes, internal audit reports, and incident records. All of these change over time, and all are signed by people. When the notified body or supervisory authority arrives, the discussion is not about whether the QMS exists. It is about whether what is presented is what was in effect on the relevant date.
A document management system can only provide that information if no one with permissions has been able to alter the history. A shared folder cannot provide that information. And a statement from the person in charge is, once again, just a statement.
An SGC under Article 17 that a third party can verify
Coverage by Requirement
- Each requirement of Article 17 and each clause of EN 18286 or ISO/IEC 42001, along with the document that addresses it, its current revision, and any open findings.
- No AI involved in the calculation: that's the record.
Signatures associated with the file
- No one signs a version other than the one in front of them: the digital signature is calculated in the signer's browser.
- Acknowledgments of receipt, approvals, and changes are given a certified date.
Sealed package for evaluation
- V-Seal™ records every audit and the complete set of documents on an immutable blockchain.
- The notified body verifies this on the public portal, without an account.
The result is an SGC whose history does not depend on anyone's word: what was in effect, who approved it, and when—all of which can be verified by anyone.
The decision regarding conformity rests with the notified body or the authority. What V-PROOF guarantees is that, when the presumption under Article 40 is invoked, compliance with the standard is supported by evidence rather than by a mere assertion.
The deadline remains December 2, 2027. What is already available is the yardstick.
You don't have to take our word for it. Any evidence sealed with V-Seal™ can be verified on the public portal—no account required—in just a few seconds.
Commission Implementing Decision (EU) 2026/2274 of October 12, 2026. See on EUR-Lex
European Commission. Standardization of the AI Act. See page
Regulation (EU) 2024/1689, Articles 17 and 40. See text
