So When Does Deactivation Become an Article 22 Decision?

Uber’s reported driver-blocking system gives a direct answer to the question at the center of this series, but the answer must be stated with legal and evidentiary care. A platform deactivation becomes the kind of decision Article 22 addresses when it is an individualized outcome, produced solely by automated processing, that has legal or similarly significant effects on the person. The Uber releases describe all three features strongly: software-linked deactivation, no human assessment before the account action, and loss of access to rides and platform income. They support a serious Article 22 concern and the regulators’ characterization, while the announced fine remains subject to appeal and the complete enforcement record is not available here.

The most useful way to carry the conclusion forward is to follow the decision path rather than the technology label. Ask what changed for the person, what actually produced that change, how serious the effect was, whether an Article 22 exception is established, and whether the process gives the person a real way to understand and challenge the result. That model applies to Uber first. It can then be used to ask disciplined questions about other high-impact platform actions without assuming that every automated warning or recommendation is legally identical.

The answer is about the decision path, not the algorithm label

The reported Uber process has a clear point of consequence. The AP says Uber used software to track driving behaviour and customer reviews between 2018 and 2022. A suspicion of fraud could lead to temporary deactivation. Low customer reviews could lead to temporary deactivation, and persistent low reviews could lead to permanent deactivation. The AP says that no human assessment occurred before these account actions. The CNIL describes the deactivations as automated individual decisions.

That description is stronger than evidence that software merely helped a person sort cases. It identifies a person-specific account outcome. Before the action, the driver could use Uber to make rides; after the action, the CNIL says, a blocked driver could no longer make rides or generate revenue. The software-linked process therefore reached the account boundary, where an internal classification became a condition of participation in the platform’s marketplace.

The first Article 22 question is consequently whether there was a decision. The relevant decision is not the customer review or the suspicion of fraud by itself. Those are inputs or classifications. The account deactivation is the outcome directed at a particular driver. Calling it a flag, hold, or status does not change what it does. A decision can be embedded in an operational system rather than issued as a formal legal document.

The second question is whether the outcome was based solely on automated processing. The AP’s case-specific statement is that the software detected the relevant condition and that there was no human assessment before the account action. That supports the conclusion that the reported path was automated in the legally important sense. It does not establish that no employee ever spoke with a driver, handled a later complaint, or restored an account. A later interaction would be a different stage of the process. It could provide a remedy without changing who made the initial deactivation decision.

The third question is the effect on the individual. The available evidence is strongest on the “similarly significantly affects” route rather than on a detailed finding about “legal effects.” The CNIL and AP both connect deactivation to the loss of rides and Uber income. The AP describes the loss as immediate. The releases do not quantify every driver’s loss or establish that all drivers relied on Uber to the same extent, but they do show why the authorities considered the consequence significant: the account state removed access to a source of work.

This distinction keeps the legal conclusion proportional to the record. It is not necessary to claim that the AP proved a formal change in a driver’s legal rights when the evidence already supports a serious practical effect. Nor is it correct to treat every service interruption as significant merely because it involves software. An internal recommendation, a low-priority ranking, and the loss of access to an income-generating platform may all be automated, but their consequences are not automatically equivalent.

The three conditions must meet in the same individual process. A platform might automatically collect a rating but leave the consequential decision to an independent assessor. It might use an automated tool to identify possible fraud but have a person examine the case before access changes. It might automate an administrative task that has no legal or serious practical effect. Those possibilities explain why Article 22 is not a general ban on algorithms. The rule follows automation when it becomes the mechanism that determines a serious outcome for a person.

On the regulators’ account, Uber’s reported deactivation path fits that model. The individualized outcome is the account block. Sole automation is supported by the absence of human assessment before the action. Significant effect is supported by the loss of rides and income. That is the strongest provisional answer to the title question.