A medtech start-up develops an AI that pre-reports X-ray images. Approval for the product is under way, and the team knows quality management from the medical device world. Then the notified body asks: how does your quality management system ensure that training data, model changes and field observations are managed under the AI Act? Since October 2026, EN 18286 is a harmonised standard that answers exactly this question.
EN 18286 specifies the requirements for the quality management system (QMS) that providers of high-risk AI systems set up under Art. 17 of the EU AI Act. The standard was developed by CEN-CENELEC JTC 21. The Commission published it as a harmonised standard in the Official Journal with Implementing Decision (EU) 2026/2274 of 12 October 2026, without restriction. This makes it the only harmonised standard under the AI Act so far.
What does the standard cover?
The standard places the elements of Art. 17(1) (points a to m) into a management system structure:
- Quality management system (section 4): identify applicable regulatory requirements, define which AI systems the QMS covers, maintain a regulatory compliance strategy, control documentation.
- Leadership and planning (sections 5 and 6): management responsibility, quality policy, roles, risks of the QMS itself, quality objectives.
- Support (section 7): resources, competence, communication with authorities, notified bodies and customers.
- Life cycle (section 8): risk management under Art. 9, requirements, verification and validation, data management, decommissioning, marking, continuously learning systems and technical documentation.
- Supply chain and operation (section 9): suppliers and model providers, changes to systems, post-market monitoring, reporting of serious incidents, nonconformities and corrective action.
- Review and change (section 10): management review and controlled changes to the QMS.
Annex ZA maps the sections to the requirements of the regulation and defines what the presumption of conformity covers.
Risks of the quality management system
Besides the risks of the AI systems, the standard considers the risks of the QMS itself. Typical scenarios:
| Scenario | What it is about |
|---|---|
| QMS misses its outcomes | Processes do not work, conformity cannot be demonstrated |
| Undesired effects | Processes tie up too many resources or are only fulfilled on paper |
| Outdated strategy | New legal acts, standards or guidelines do not reach procedures and documentation in time |
| Technology overwhelms the processes | New model types, continuously learning systems or third-party models slip through review and release steps |
| Missing information from third parties | Suppliers or model providers deliver information for documentation and assessment late or not at all |
| Missing resources | Not enough people with the necessary competence for review, monitoring or documentation |
| Outage of the QMS software | Documents are unavailable, including to the authority |
Relationship to ISO 42001 and the AI Act
ISO/IEC 42001 remains the management system for AI, for every role. EN 18286 is added for providers of high-risk AI; the two standards are additive. An existing AI management system provides context, leadership, risk assessment, internal audit and management review; EN 18286 aligns these parts with the obligations of the AI Act and adds the life cycle, data, post-market monitoring and incident reporting.
The presumption of conformity has limits. It covers Art. 17(1) and the first sentence of Art. 11(1), to the extent assigned by Annex ZA. The first sentence of Art. 11(1) only applies with technical documentation containing all the information in Annex IV. Art. 17(2) to (4) lie outside the standard, as do further provider obligations such as Art. 9, 10, 12 to 15, 43, 47 to 49, 72 and 73. Since the Digital Omnibus, the AI Act provides a simplified QMS for small and medium-sized enterprises (Art. 63).
Free template
The AI provider policy from the AI Compliance Starter Kit is a quality management system under Art. 17 that follows EN 18286: all thirteen aspects, process table, life cycle with entry and exit criteria and allocation of responsibilities. The quality policy under clause 5.2 of the standard is part of the AI policy.
What Cenedril covers
Cenedril’s AI compliance assistant supports providers of high-risk AI systems in building and running a quality management system under Article 17 of the AI Act that reflects the requirements of the harmonised standard EN 18286:2026 (Implementing Decision (EU) 2026/2274, OJ L of 14.10.2026). Procedures, records and evidence are created in the assistant; the provider obtains the presumption of conformity under Article 40 by applying the standard in its organisation.
This includes, among other things, the provider policy with the compliance strategy, the approach per requirement in the technical documentation, the risks of the QMS in the AI risk register, workflows for development and release, changes and supplier assessment, post-market monitoring, incidents and corrective actions, and the management review with the evaluations for providers. Cenedril does not issue certificates and replaces neither the provider’s own assessment nor a notified body.
Related standards and laws
- EU AI Act: the regulation that Art. 17 comes from.
- ISO/IEC 42001: the AI management system that EN 18286 builds on.
- ISO 31000: principles of risk management.
- ISO 19011: guidelines for auditing management systems.
Sources
- Implementing Decision (EU) 2026/2274 (EUR-Lex): harmonised standards for high-risk AI systems, Official Journal of 14 October 2026
- Regulation (EU) 2024/1689 (EUR-Lex): AI Act, Art. 17 and Art. 40
- CEN-CENELEC JTC 21: standardisation on artificial intelligence