EN 18286, the first EU AI Act harmonised standard, for companies that use AI

EN 18286 is the standard for the quality management system that Article 17 requires from providers of high-risk AI systems. Article 17 lists what that system must contain in writing, among them a strategy for regulatory compliance, design and development controls, testing and validation, data management, the risk management system, post-market monitoring, serious-incident reporting, record-keeping and an accountability framework.

The text of the standard itself is sold by the national standards bodies. What is public is its title, its scope as described in the Commission's decision, and JTC 21's summary: requirements and guidance for a quality management system, with annexes that map it to ISO 9001 and ISO/IEC 42001, so an organisation that already runs either can build on it.

The presumption reaches only as far as the standard does. Article 40 limits it to the requirements a standard covers, and EN 18286 covers quality management. It says nothing about whether a given system is accurate, robust or secure enough.

Providers develop an AI system, or have it developed, and place it on the market under their own name. A provider of a high-risk system needs a quality management system, and EN 18286 is one way to build it.

Deployers use an AI system under their authority. A company that gives staff Claude or GPT for drafting, coding or summarising, or builds an internal assistant on an LLM API, is a deployer of those tools. If a use is high-risk, Article 26 applies: use the system according to its instructions for use, assign human oversight, monitor it and tell the provider about risks and serious incidents, and keep the logs the system generates under your control for at least six months.

When a deployer becomes a provider. Article 25 treats a deployer as the provider of a high-risk system if it puts its name or trademark on one, makes a substantial modification to one, or changes the intended purpose of an AI system, including a general-purpose AI system, so that it becomes high-risk. A company that builds a CV screening or credit scoring tool on a general-purpose model should check this line first. If it applies, Article 17 and the quality management system are the company's own obligation.

When you buy an AI system for a use that may be high-risk, or a model or service that a high-risk system will be built on, these questions turn the standard into procurement terms. Copy them into your RFP template or vendor questionnaire.

EN 18286 is the first of the harmonised standards the Commission asked for, and the only one cited so far. CEN-CENELEC JTC 21 is still working on standards for risk management, data governance and quality, logging, transparency, human oversight, accuracy, robustness, cybersecurity and conformity assessment. It says several core drafts are being revised after their public enquiries. Each will go through adoption by CEN and CENELEC, assessment by the Commission and citation in the Official Journal before it gives a presumption of conformity.

For procurement this means the question set above will grow. Ask vendors to tell you when they apply a newly cited standard, and review the answers before the high-risk rules apply to Annex III systems on 2 December 2027.

Whatever your vendors answer, a deployer needs its own record of which models and providers each application uses. Standard Compute is one way to keep that record for API use. Every admin change is recorded with before and after values, and you can export a record of every request showing the model, the provider that served it and the region rule in force, kept for 12 months.

Decide which models each key can use, so every app and every team member gets exactly the models you approve. Choose exactly which inference providers may handle your data, for the whole company, a team or a single key. Set a Europe rule, and every request, including retries and fallbacks, is served only by approved European inference endpoints.

These are gateway records of model and provider use. The logs Article 26(6) refers to are the ones a high-risk system generates itself. Your legal team decides what the AI Act requires of your company.