The words that decide what you owe

Four words in the EU AI Act decide which duties land on you. Using them loosely is how organisations end up discovering, late, that they are responsible for something they assumed was the vendor's problem.

Provider and deployer

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 trade mark. A deployer uses an AI system under its own authority, in the course of a professional activity.

Most organisations are deployers. That is a lighter set of duties, but it is not nothing: a deployer must use the system according to its instructions, assign human oversight to someone with the competence, training and authority to exercise it, keep the automatic logs, and inform workers before an AI system is used on them at work.

How a deployer becomes a provider without noticing

This is the part worth remembering, because it is the mechanism by which the heavy obligations arrive unannounced. Under Article 25, a deployer becomes a provider, with the full provider obligation set, in three situations:

• You put your name or trade mark on it. Rebranding a bought system as your own makes it yours in law.
• You make a substantial modification. Meaningful fine-tuning is the common route.
• You change the intended purpose so that the system becomes high-risk. Using a general summariser to screen job applications is the textbook example: the tool did not change, the purpose did.

Nobody sends a notification when this happens. It is discovered later, usually by someone assembling a technical file that does not exist.

AI system and general-purpose AI model

The Act distinguishes an AI system, the thing that is used, from a general-purpose AI model, the underlying model that can serve many purposes. A large language model is a GPAI model; the assistant your colleagues open in a browser is an AI system built on one.

The distinction has practical force. GPAI obligations sit with whoever put the model on the market, and have applied since 2 August 2025. Your duties concern the system you built or use, not the model underneath, unless you did something to the model that made it yours.

Intended purpose

The single most consequential phrase in the Act, and the one people treat as marketing copy. The intended purpose is the use for which a system is intended by the provider, as stated in the instructions for use. It determines the risk classification, which determines everything else.

Two consequences follow. First, using a system outside its intended purpose puts you outside the assurances that came with it, and can make you the provider. Second, a vendor who writes a vague intended purpose is transferring risk to you: "general business productivity" tells you nothing about whether the tool may be used to screen candidates.

The vocabulary you will hear that is not in the Act

"Algorithm" appears constantly in discussion and decides nothing. "Machine learning" is a technique, not a legal category. "Automated decision-making" is a data-protection concept rather than an AI Act one: it appears in GDPR Article 22 and in Article 21 of the revised Swiss Federal Act on Data Protection, and it can bind you even where the AI Act does not.

Keep the two vocabularies separate. A system can be outside the AI Act's high-risk list and still be a solely automated decision with legal effect, which brings its own duties to inform, to allow a view to be stated, and to obtain human review.

As of 24 September 2026.