Provider or deployer? The line that decides your obligations

Edition 09. A reading time of about 5 minutes.

The hook

Almost everyone who uses AI in hiring is a "deployer," and that is the comfortable side to be on. The deployer's duties, the ones these editions have walked through, are real but manageable: oversight, monitoring, records, the right vendor questions. The "provider" carries a much heavier load: the technical documentation, the conformity assessment, the deep build work. The catch is that three ordinary-looking moves can flip you from deployer to provider without anyone announcing it. Knowing where the line sits is how you avoid stepping over it by accident.

What this means

By default you are a deployer: you use a system someone else built and placed on the market. But Article 25(1) names three ways a deployer, distributor, importer, or any third party becomes the provider of a high-risk system, and inherits the provider's obligations (AI Act, Article 25):

- You put your own name or trademark on a high-risk system already on the market. - You make a substantial modification to a high-risk system, and it stays high-risk. - You change the intended purpose of a system, even a general-purpose or non-high-risk one, so that it becomes high-risk.

For hiring, the last two are the quiet traps. Customize a hiring tool deeply enough, or take a general-purpose model and point it at evaluating candidates, and you may have changed the system in the eyes of the Act, and stepped into the provider's far larger set of duties. There is a sting in the tail, too: when you cross that line, the original provider is no longer considered the provider of that system (AI Act, Article 25). They must still cooperate and share information, but the obligation, and the exposure, has moved to you. There is no longer anyone upstream to point at.

The reassuring half: none of this happens by merely using a tool as sold, within its intended purpose. Deployment, as intended, keeps you a deployer.

What to do with this

Stay a deployer on purpose, and notice when you are being tempted off it. Four checks.

1. Use tools as intended. Operating a hiring system the way the provider designed it, within its stated purpose, keeps you on the deployer side. That is the whole point of the instructions for use.
2. Treat heavy customization as a flag, not a feature. If you are modifying a tool substantially, ask the vendor in writing whether that modification makes you the provider. The answer changes your obligations entirely.
3. Be careful pointing general-purpose tools at hiring. Using a general model to screen or score candidates can change its intended purpose into a high-risk one, and make you its provider. Decide that deliberately, not by drift.
4. Do not put your brand on someone else's system lightly. Rebranding a high-risk tool as your own is, on its own, enough to make you the provider.

The line is not exotic, and most teams will never cross it. The ones who do almost always cross it by accident, through a customization or a clever reuse that nobody flagged. One question at the right moment keeps the heavy obligations with the people who actually signed up for them.

This is educational material and a starting point, not legal advice. The EU AI Act is still being amended, and the high-risk timeline rests on an agreement that is not yet final law. For whether a specific customization or reuse makes you a provider under Article 25, consult a qualified lawyer. Signato is not a law firm and does not certify compliance.

AI that drafts at scale can also send the wrong thing at scale. See the Signato chamber block a forbidden send before it happens: run the free local trial, no card, nothing leaves your machine.