Provider or Deployer? Which Article 50 Obligation Applies to You

Almost every practical Article 50 question resolves into this one first, and answering it in the wrong order wastes the rest of the work. A team that reads Article 50(2), concludes it needs to embed machine-readable marking, and starts scoping a watermarking project has usually just adopted a duty that belongs to its tool vendor.

The role decides the obligation, so settle it first

Article 50 splits along the provider and deployer line, and the two halves have almost nothing in common:

The roles attach to an AI system, not to a company. The same organisation can be a provider of the feature it sells and a deployer of the twenty tools it buys, and the answer has to be worked out for each.

What makes you a provider under Article 3(3)

Two limbs, and both have to hold:

“Whether for payment or free of charge” closes the obvious gap. Shipping a free tool does not move you out of the definition.

The phrase carrying the most weight is “has an AI system developed”. Commission a bespoke system from an agency, put your brand on it and release it, and the first limb is met even though nobody in your company wrote any of it. Buying a seat on a system somebody else operates and markets is a different thing entirely.

Article 3(68) adds a related role for anyone building on top of a model. A downstream provider is a provider of an AI system that integrates an AI model, whether that model is their own or another party’s. If you wrap a third-party model in your own product and sell it under your name, that is the definition you are reading.

What makes you a deployer under Article 3(4)

This is a wide definition and it is meant to be. “Using an AI system under its authority” covers a marketing team generating campaign images, an agency doing it for a client, a retailer producing product visuals, and a newsroom drafting copy. No technical involvement is required and no contract with the model vendor beyond ordinary use.

The Commission’s FAQ describes deployer use as professional activity, including activity that provides an economic benefit. That framing matters more than the word “business”: a sole trader, a freelancer and a non-profit publishing to promote its work are all engaged in something other than a personal non-professional activity.

Can a deployer become a provider?

Yes, but not by the route most commentary cites.

Article 25 is about high-risk systems, and Article 50 systems generally are not high-risk. Article 25(1) converts a distributor, importer, deployer or third party into “a provider of a high-risk AI system” where they put their name on one, substantially modify one, or modify a system’s intended purpose so that it becomes high-risk. Every limb is tied to high-risk classification under Article 6. A generative image tool used for marketing is a transparency-tier system, so Article 25 is not the provision that moves you.

The route that does is Article 3(3) itself. Take a model, build a system around it, and put that system on the market under your own name, and you satisfy both limbs of the definition directly. You do not need Article 25 to get there, and Article 25 will not get you there for a system that never becomes high-risk.

Does the AI Act reach you if you are outside the EU?

Article 2(1) sets the territorial scope, and it is deliberately broad. It catches providers placing AI systems on the Union market irrespective of where they are established, deployers established or located within the Union, and providers and deployers located in a third country where the output produced by the system is used in the Union. Importers, distributors, product manufacturers and authorised representatives are named alongside them.

The third limb is the one that surprises people. It attaches to where the output lands rather than to where you sit. How far it stretches for published marketing content viewed by an EU audience has not been tested, and anyone relying on being outside the Union should treat that as an open question rather than a settled answer.

The personal, non-professional carve-out is narrower than it sounds

Article 3(4) excludes AI systems used “in the course of a personal non-professional activity” from the deployer definition. Article 2(10) does similar work at the level of scope, disapplying the Regulation to the obligations of deployers who are natural persons using AI systems in the course of a purely personal non-professional activity. Generating images for a family album is outside. Generating them for a side business, a monetised channel or a client is not, on the Commission’s reading that professional activity includes activity providing an economic benefit.

Two things the carve-out does not do. It does not follow the content: an image made privately and later published commercially is being deployed by whoever publishes it. And it says nothing about Article 50(2), which binds the provider regardless of what any individual user does with the tool.

Open source is not an exemption from Article 50

The transparency obligations are named in the carve-out from the carve-out. A provider releasing an open-weights generative model still reads Article 50(2), and a deployer using one still reads Article 50(4). This catches teams that moved to self-hosted open models partly on the understanding that it reduced regulatory exposure, and on this particular provision it does not.

Article 53(2) does grant open-source general-purpose AI model providers relief from the documentation duties in Article 53(1)(a) and (b), unless the model carries systemic risk. That is a different chapter and a different obligation, and it does not reach into Article 50.

What each role has to do under Article 50

For a provider: build the disclosure into the system. Article 50(1) is a design requirement for systems that interact with people, and Article 50(2) is a marking requirement for synthetic output. Systems placed on the market before 2 August 2026 have until 2 December 2026 for the Article 50(2) marking duty alone.

For a deployer: decide, per publication, whether the content meets the Article 50(4) trigger, and disclose where it does. The Commission has stated that a deployer cannot discharge this by pointing at the provider’s machine-readable marking, because the disclosure has to be perceivable by a person without special tools. That distinction is worked through in Visible Label or Machine-Readable Marking?, and the deepfake trigger itself in When Does an AI-Generated Image Need a Visible Label Under Article 50?.

If you hold both roles, the two sets of obligations run in parallel rather than merging. Discharging one has no bearing on the other.