First harmonised standard under the EU AI Act published: de facto legal standard for AI-specific quality management
By publishing the reference to the European standard EN 18286:2026, the European Commission has established the first harmonised standard within the meaning of the EU AI Act. In considerably greater detail than the AI Act itself, the standard offers providers of high-risk AI systems (or AI models) a blueprint on how to design a quality management system (QMS) in accordance with Article 17 of the AI Act. Whilst its application is not legally mandatory, compliance offers providers a significant advantage: a presumption of conformity, which effectively makes this standard the de facto legal standard for the QMS of high-risk AI systems.
Background: How did the standard come about?
After three years of work, the European standardisation organisations – namely the European Committee for Standardisation (CEN) and the European Committee for Electrotechnical Standardisation (Cenelec) – developed the European standard EN 18286:2026 ‘Artificial Intelligence – Quality Management System for Regulatory Purposes within the Framework of the AI Act’ on behalf of the European Commission and adopted it in July 2026. The European Commission subsequently reviewed the standard and concluded that it reflects the requirements of the EU AI Act regarding quality management systems. It has therefore now published the reference to the standard in Implementing Decision (EU) 2026/2274 of 12 October 2026. Pursuant to Article 40 of the AI Act, this means that the European standard EN 18286:2026 is the first so-called harmonised standard within the meaning of the AI Act.
What does the standard cover?
EN 18286:2026 serves as a practical manual for providers of high-risk AI systems. It describes in detail how to build a quality management system that meets the requirements of Article 17 of the AI Act. The standard covers the entire lifecycle of an AI system, from design, development, verification and validation, through operation and modification, to decommissioning. It sets out requirements to ensure that each phase can be organised in a manner that supports legal compliance and traceability. This includes supporting processes such as data management and labelling. Ultimately, EN 18286:2026 offers guidance on how a provider organises compliance with the AI Act. It does not, however, determine which characteristics the AI system must possess.
Under the AI Act, the remaining requirements, namely those concerning risk management, data governance, transparency, human oversight, accuracy, robustness and cybersecurity, are also to be addressed by harmonised standards. However, these standards have yet to be published.
Relationship to other QMS standards and existing systems
Many organisations already operate established management systems based on standards such as ISO 9001 or ISO/IEC 42001. EN 18286:2026 is not intended to be a replacement, but rather a supplemental framework that specifically addresses the requirements of the EU AI Act. A separate quality management system for the AI Act is not required. Annexes to EN 18286:2026 ‘map’ the content of the standard to ISO 9001 and ISO/IEC 42001, providing guidance for organisations with existing management systems.
Why is it worth adopting?
The application of EN 18286:2026 is voluntary rather than legally mandatory. Nevertheless, there are two compelling reasons for providers of high-risk AI systems and general-purpose AI models to adopt it: it serves as a helpful blueprint for a quality management system and establishes a presumption of conformity.
- Blueprint. Article 17 of the AI Act lists thirteen minimum requirements for a quality management system but provides little guidance on how to implement them. Providers have therefore so far been left with the question of what a system in compliance with the AI Act must look like. EN 18286:2026 answers this question. It translates the abstract requirements into a structured framework of processes, roles and documentation requirements, against which the system’s lifecycle can be aligned. Providers therefore do not have to design their own quality management system, which saves time and money. At the same time, the standard establishes a common benchmark for testing bodies, customers and business partners. Providers who can demonstrate compliance with the standard gain an advantage, for example in tenders or contract negotiations.
- Presumption of conformity. Those who apply the standard also benefit from a legal perspective: if a provider’s quality management system complies with EN 18286:2026, it is presumed to meet the requirements of Article 17 of the AI Act (known as the ‘presumption of conformity’). A third party would first have to prove that the requirements of Article 17 of the AI Act are not met. This significantly increases legal certainty for AI providers.
The benefits are particularly evident for companies that already operate a quality management system. Certification to ISO 9001, ISO/IEC 42001 or ISO 13485 alone does not establish a presumption of conformity. However, the existing processes can remain in use. By supplementing them with the AI-specific requirements of EN 18286:2026, providers can achieve the presumption of conformity without having to set up a second system. The manageable effort required for these adjustments is thus outweighed by a considerable legal advantage.
Anyone opting for their own approach forgoes these advantages. Under Article 17(1)(e) of the AI Act, they must set out the alternative means by which they meet the requirements and justify why their solution is equivalent to the provisions of EN 18286:2026. The risks, costs and effort involved are unlikely to be proportionate to the potential benefits. In practice, EN 18286:2026 is therefore likely to become the de facto legal standard for the QMS.
Conclusion
Article 17 of the AI Act applies from 2 December 2027 to high-risk AI systems as defined in Annex III and from 2 August 2028 to those defined in Annex I. Establishing a quality management system takes time. Providers of high-risk AI systems should therefore begin aligning their quality management with EN 18286:2026 without delay.