Provider vs Deployer Under the EU AI Act: Which One Are You?

A provider develops an AI system (or has one developed) and places it on the market or puts it into service under its own name or trademark — whether paid or free. A deployer uses an AI system under its own authority in a professional context. Personal, non-professional use is out of scope entirely.

The roles are per system, not per company. A SaaS company is provider of the AI features it ships, and simultaneously deployer of the AI tools it uses internally — the hiring platform it sells and the resume screener it uses on its own candidates carry different hats.

The provider stack for high-risk systems: risk management system (Art. 9), data governance (Art. 10), Annex IV technical documentation (Art. 11), automatic logging capability (Art. 12), transparent instructions for use (Art. 13), human-oversight-by-design (Art. 14), accuracy, robustness and cybersecurity (Art. 15), quality management system (Art. 17), conformity assessment and CE marking (Art. 43, 47–49), EU database registration, post-market monitoring (Art. 72) and serious-incident reporting (Art. 73).

Non-EU providers with EU customers additionally need an authorised representative established in the Union (Art. 22) — a frequently missed line item for US and Asian SaaS.

Deployers of high-risk systems must use them per the provider’s instructions, assign competent human oversight, ensure input data is relevant and representative, monitor operation, keep logs at least six months, inform workers’ representatives before workplace deployment, inform affected persons subject to the system’s decisions, and report serious incidents (Art. 26). Article 27 FRIA applies only to defined deployers and use cases: public-law bodies, private entities providing public services, and Annex III 5(b)/(c) creditworthiness or insurance systems.

Deployers of limited-risk systems mostly inherit the transparency reality: make sure the disclosure the provider built actually reaches the end user. And every professional user of AI carries Article 4 AI-literacy duties for staff.

Article 25 converts a deployer (or distributor, or any third party) into the provider of a high-risk system when they: (1) put their name or trademark on a high-risk system already on the market — the white-label trap; (2) make a substantial modification to it — the fine-tuning-and-repackaging trap; or (3) modify the intended purpose of a system so that it becomes high-risk — the repurposing trap, e.g. wiring a general chatbot into candidate screening.

In all three cases the full provider stack transfers. The original provider must hand over documentation and cooperate, but the obligations are now yours. If your product is a branded wrapper around a third-party model doing Annex III work, budget as a provider, not a deployer.

Expect the AI Act to show up in your enterprise contracts from both directions. Your EU customers, as deployers, will ask you for the artifacts that let them comply: instructions for use, oversight guidance, logging access, incident-notification terms. And if you consume upstream models, you need the mirror-image terms from your model providers.

The vendor who arrives with a provider pack ready — classification memo, Annex IV documentation, model card, instructions for use — turns this from a negotiation blocker into a checkbox.