Short answer
EN 18286:2026 is the first harmonised standard under the EU AI Act: Commission Implementing Decision (EU) 2026/2274 cited it in the Official Journal on 14 October 2026. It turns the quality management system required by Article 17 of the AI Act into auditable requirements. It is sector-neutral, not a medical device standard, but it matters for AI medical devices that need a Notified Body, because they are high-risk AI systems. You do not need a second QMS: Article 17(3) lets you add the AI Act elements to your ISO 13485 system. Using the standard is voluntary, and the high-risk obligations apply to AI medical devices from 2 August 2028.
Key facts
- What: EN 18286:2026, “Artificial intelligence – Quality management system for EU AI Act regulatory purposes”, from CEN/CENELEC JTC 21.
- Status: harmonised under the AI Act since 14 October 2026 (Commission Implementing Decision (EU) 2026/2274). It is the first harmonised standard under the AI Act.
- Is it medical? No, it is sector-neutral. But AI medical devices that need a notified body are high-risk AI systems, so it applies to them.
- From when: the AI Act high-risk obligations, including the QMS, apply to AI medical devices from 2 August 2028. Two open points could still narrow this; see what could still change.
- ISO 13485: EN 18286 does not replace it. You extend your ISO 13485 QMS to cover the AI Act, as Article 17(3) allows.
What EN 18286 is
Article 17 of the AI Act requires providers of high-risk AI systems to have a quality management system. It lists thirteen aspects that the system must cover, from a strategy for regulatory compliance to data management, post-market monitoring and serious incident reporting. EN 18286 turns these aspects into auditable QMS requirements.
The standard was developed by CEN/CENELEC Joint Technical Committee 21 (Artificial Intelligence) under the Commission’s standardisation request for the AI Act, and published in July 2026. It has 51 pages. On 14 October 2026, Commission Implementing Decision (EU) 2026/2274 published its reference in the Official Journal. The Decision lists only this standard, with no restrictions.
From that date, a provider that follows EN 18286 benefits from a presumption of conformity with the AI Act requirements the standard covers. Annex ZA of the standard shows which requirements those are. The presumption is voluntary. You can meet Article 17 another way, but then you must show compliance yourself.
Two points set it apart from familiar management system standards:
- It is written for the AI Act, not for general good practice. Its title says “for EU AI Act regulatory purposes”, and it expects you to identify regulatory requirements and plan how to meet them.
- It is about the product, not only the organisation. Clauses 8 and 9 follow the AI system through design, data, verification, release, changes and post-market monitoring, much as ISO 13485 follows a medical device.
Does it apply to medical devices?
EN 18286 is not specific to any sector, and it is not cited under the MDR or IVDR. Whether it matters to you depends on the AI Act status of your device.
Under Article 6(1) of the AI Act, an AI system is high-risk if it is a product, or a safety component of a product, covered by the legislation in Annex I, and that product needs a third-party conformity assessment. The MDR and IVDR are in Annex I, Section A, and they stayed there after the Digital Omnibus. So in practice, an AI medical device that needs a notified body is a high-risk AI system, and Article 17 applies to its manufacturer.
| Device | Notified body involved? | High-risk AI under Article 6(1)? |
|---|---|---|
| MDR Class IIa, IIb or III | Yes | Yes, if the AI is the device or a safety component of it |
| MDR Class I with a sterile, measuring or reusable surgical aspect | Yes, for that aspect | Yes, per MDCG 2025-6 |
| MDR Class I, self-declared | No | No, unless the use is listed in Annex III (for example emergency patient triage) |
| IVDR Class B, C or D, or Class A sterile | Yes | Yes, if the AI is the device or a safety component of it |
| IVDR Class A, non-sterile | No | No |
| In-house device under MDR/IVDR Article 5(5) | No | No, under Article 6(1) |
Two cautions. First, the AI Act does not move a device into a higher MDR or IVDR class. Second, the Digital Omnibus added paragraphs 1a to 1c to Article 6, which clarify when an AI component counts as a safety component. Assess each AI function, not the product label, and record the rationale. If the device itself is not yet classified, start with MDR Rule 11.
The standard’s own introduction makes the link to medical devices. It notes that medical device manufacturers commonly work to EN ISO 13485, and recommends building the EN 18286 requirements into the existing processes rather than running a separate QMS.
Dates
| Date | What happens |
|---|---|
| July 2026 | EN 18286:2026 published by CEN and CENELEC |
| 27 July 2026 | Digital Omnibus on AI (Regulation (EU) 2026/1744) in force |
| 14 October 2026 | Implementing Decision (EU) 2026/2274 published in the Official Journal: EN 18286 becomes a harmonised standard under the AI Act |
| January 2027 | Latest date for national standards bodies to adopt EN 18286 and withdraw conflicting national standards |
| 2 December 2027 | High-risk obligations apply to Annex III systems (stand-alone uses, including emergency patient triage) |
| 28 January 2028 | Deadline in the amended Article 43(3) for notified bodies under the Annex I Section A legislation to apply for designation under the AI Act |
| 2 August 2028 | High-risk obligations, including the Article 17 QMS, apply to AI medical devices under Annex I |
The Digital Omnibus set fixed dates for the high-risk obligations. Publication of EN 18286 does not bring any of them forward.
For devices placed on the market before 2 August 2028, Article 111(2) applies the high-risk obligations only if their design changes significantly afterwards. For software that is released continuously, discuss early with your notified body what counts as a significant change. The full timeline, including what already applies, is in our EU AI Act guide for medical software.
What is in the standard
EN 18286 uses the familiar layout of management system standards, with clauses 4 to 10 and informative annexes. The table summarises the published table of contents.
| Clause | Title | What it is about |
|---|---|---|
| 4 | Quality management system | Identifying regulatory requirements, the scope of the QMS, a strategy for regulatory compliance, documented information |
| 5 | Leadership | Quality policy; roles, responsibilities and authorities |
| 6 | Planning | Risks to the functioning of the QMS; quality objectives |
| 7 | Support | Resources, competence, communication, awareness |
| 8 | AI system realization | Lifecycle processes, risks of AI systems, design and development, verification and validation, data management, retirement, identification, continuous learning AI systems, product documentation |
| 9 | Operations and control | Placing on the market, support services, supply chain, modifications, post-market monitoring, reporting serious incidents, non-compliance |
| 10 | Performance evaluation | Management review, planning of changes |
| Annex A | Interested parties | Needs and expectations regarding health, safety and fundamental rights |
| Annex B and C | Correspondence tables | Clause correspondence with EN ISO 9001:2015 and EN ISO/IEC 42001:2026 |
| Annex ZA | Relationship with the AI Act | Which requirements of Regulation (EU) 2024/1689 the standard aims to cover |
Note what the annexes map to. Annex B maps to ISO 9001 and Annex C to ISO/IEC 42001. There is no mapping to ISO 13485, so medical device manufacturers have to do that mapping themselves.
EN 18286 and your ISO 13485 QMS
Article 17(3) of the AI Act lets manufacturers with a QMS under sectoral law, such as the MDR or IVDR, include the AI Act aspects in that system. MDCG 2025-6, the joint guidance of the MDCG and the AI Board, describes the AI Act requirements as complementary to the MDR or IVDR QMS and confirms they may be included in it. For technical documentation, Article 11(2) of the AI Act asks for a single set. In practice: one QMS and one technical file, extended where the AI Act adds something new.
An ISO 13485 QMS for software already covers much of Article 17: design control, verification and validation, document and record control, purchasing, feedback and reporting to authorities. The gaps are mostly about data, models and fundamental rights. The table below is a starting point for the gap assessment.
| Article 17(1) | EN 18286 clause (by title) | ISO 13485 starting point | What SaMD teams usually need to add |
|---|---|---|---|
| (a) Strategy for regulatory compliance, incl. conformity assessment and management of modifications | 4.2, 4.4, 9.4 | 4.1.1, 7.3.9 | The AI Act in the regulatory requirements register; the conformity route through your notified body; a rule for which model changes are significant |
| (b), (c) Design, development, quality control and assurance | 8.1, 8.3 | 7.3, IEC 62304 | AI-specific design inputs: performance metrics, human oversight, transparency, logging |
| (d) Examination, test and validation | 8.4 | 7.3.6, 7.3.7 | Model verification on independent test data, performance in relevant subgroups, robustness |
| (e) Technical specifications, including standards | 4.2 | 7.3.3 | A list of AI standards and common specifications that you apply, updated as more are cited |
| (f) Data management | 8.5 | No direct equivalent | The biggest gap: data acquisition, labelling, splits, provenance, bias examination and retention (Article 10) |
| (g) Risk management (Article 9) | 8.2 | 7.1 with ISO 14971 | Fundamental rights alongside health and safety; AI hazards such as bias, drift and over-reliance on outputs |
| (h) Post-market monitoring (Article 72) | 9.5 | 8.2.1 and the MDR PMS system | Performance monitoring in use, including drift, inside the same PMS plan |
| (i) Serious incident reporting (Article 73) | 9.6 | 8.2.3 and MDR vigilance | A decision rule for incidents that infringe fundamental rights; other serious incidents stay in MDR or IVDR vigilance (Article 73(10)) |
| (j) Communication with authorities, notified bodies, operators and customers | 7.3 | 5.5.3, 7.2.3 | Information for deployers; handling authority requests under the AI Act |
| (k) Record-keeping | 4.5, 8.9 | 4.2.4, 4.2.5 | Retention of 10 years after placing on the market (Article 18), aligned with your MDR retention rule |
| (l) Resource management, incl. security of supply | 7.1, 9.3 | 6, 7.4 | Datasets, pre-trained models and compute providers treated as suppliers |
| (m) Accountability framework | 5.3 | 5.5.1, 5.5.2 | Named owners for data, models and human oversight |
For most SaMD teams, three areas need the most new work:
- Data management. ISO 13485 has no requirements for training, validation and test data. You need a procedure for how data are collected, labelled, split, checked for bias and kept, and records that show it was followed for each model version.
- Model changes. Retraining, new data and threshold changes need rules for when a change is significant, for MDR and AI Act purposes. See AI/ML SaMD change control.
- Risk beyond health and safety. ISO 14971 deals with harm to health. The AI Act adds fundamental rights, such as non-discrimination. Add these to the same risk process rather than starting a separate one. See ISO 14971 for medical software.
Serious incidents are simpler than they look. For AI systems that are medical devices or safety components of them, Article 73(10) limits AI Act incident notification to infringements of fundamental rights obligations. Other serious incidents continue to go through MDR or IVDR vigilance. What you need is a decision rule in your vigilance procedure, not a second reporting process.
Notified bodies and certificates
For AI medical devices, the conformity assessment stays under the MDR or IVDR. The Digital Omnibus replaced Article 43(3) of the AI Act:
- the manufacturer follows the conformity assessment procedure of the MDR or IVDR;
- notified bodies designated under that legislation can assess the AI Act requirements, after a check of certain AI Act criteria;
- the assessment of the Article 17 QMS is part of it;
- these notified bodies have until 28 January 2028 to apply for designation under the AI Act.
So there is no separate “EN 18286 certificate” that the law asks for. The question to put to your notified body now is whether it plans to apply for AI Act designation, and how it will assess the AI Act QMS elements in your next audits.
Simplified QMS for SMEs
The Digital Omnibus made two changes that matter for small manufacturers:
- Proportionality (Article 17(2)). The QMS must be proportionate to the size of the organisation, and the amended text names SMEs, including start-ups, and small mid-caps.
- Simplified elements (Article 63(1)). SMEs, including start-ups, without partner or linked enterprises, may comply with certain elements of the QMS in a simplified way. Before the Omnibus, this applied only to microenterprises. The Commission is to publish guidelines on which elements.
These changes concern the AI Act part of the QMS. They do not reduce the MDR or IVDR quality management requirements, which still apply in full.
What could still change
Two open points could reduce how much of Article 17 applies directly to medical devices:
- Delegated acts under Article 2(13). The Digital Omnibus allows the Commission to limit specific AI Act requirements, including Article 17, for Annex I Section A products such as medical devices, where the sectoral legislation already gives an equivalent or higher level of protection. The acts are due by 2 August 2027. Which requirements will be limited for MDR and IVDR devices, if any, is not yet known.
- The MDR/IVDR revision proposal COM(2025) 1023. It would move the MDR and IVDR to Section B of Annex I. Only a few AI Act provisions apply directly to Section B products, and Article 17 is not one of them; the AI requirements would reach devices through acts under the MDR and IVDR instead. This is a proposal under negotiation, not law.
Neither changes the practical advice. Control of training data, model changes and AI performance in use is expected under the MDR as well, through software lifecycle, risk management and post-market surveillance. Work on these is not wasted, whichever way the open points go.
What to do now
- Confirm your AI Act status per AI function. Use Article 6(1), the new paragraphs 1a to 1c and MDCG 2025-6, and record the rationale in your regulatory strategy.
- Buy EN 18286. It is sold through national standards bodies, and national editions must be available by January 2027.
- Run a gap assessment against your ISO 13485 QMS. Go clause by clause, using Annex ZA and the table above. Plan to extend existing procedures, not to write a parallel set.
- Start with the long lead items. Data management, model change control, performance monitoring in use, and the incident decision rule need records over time, so they cannot be written the month before an audit.
- Talk to your notified body. Ask about its AI Act designation plans and how the AI Act elements will be audited. Align this with your certificate renewal and surveillance audit dates before 2 August 2028.
- Track further standards. JTC 21 is revising further drafts, including risk management, data, logging, accuracy, robustness and cybersecurity. Update your list of applied standards as they are cited.
If you are building or extending a QMS for AI software, this is the core of our ISO 13485 QMS implementation service. For a review of where your current QMS and technical file stand against the AI Act, see the technical documentation gap analysis.
Sources
- Commission Implementing Decision (EU) 2026/2274 of 12 October 2026 on the harmonised standards for high-risk AI systems (OJ L, 14.10.2026) — cites EN 18286:2026
- CEN-CENELEC: EN 18286:2026 published in the OJEU (15 October 2026)
- CEN-CENELEC JTC 21: EN 18286 becomes the first harmonised standard under the EU AI Act (14 October 2026) — presumption for Article 17; upcoming standards
- EN 18286:2026 – sample pages incl. table of contents and introduction (iTeh Standards) — clauses, annexes, national adoption by January 2027
- Regulation (EU) 2024/1689 (AI Act), consolidated text of 27 July 2026 — Articles 6, 17, 18, 43, 63, 73, 111, 113
- Regulation (EU) 2026/1744 (Digital Omnibus on AI) — amended Articles 2, 6, 17(2), 43(3), 63(1), 113
- Proposal for a regulation to simplify rules on medical and in vitro diagnostic devices, COM(2025) 1023
- MDCG 2025-6 / AIB 2025-1 – FAQ on the interplay between the MDR/IVDR and the AI Act (June 2025) — Table 1, Q6, Q12, Q28