Dutch regulator fines Uber 825 million euros for switching off drivers by computer

Tech and AI

Dutch regulator fines Uber 825 million euros for switching off drivers by computer

By Staff Writer  |  23 August 2026

A distant view of the Amsterdam skyline at night, lit buildings low on the horizon under a dark sky

The penalty is the second largest yet issued under the European data protection regulation. The finding is that accounts were deactivated with no human looking at the decision.

The Dutch data protection authority has fined Uber 824,990,000 euros for taking fully automated decisions about drivers. Software watched driving behaviour and customer ratings. Where it read a pattern as fraud, or read the ratings as too low, the driver's account was switched off. Where the low ratings persisted, the account was switched off permanently. No person assessed any of it, and the regulator found that drivers were not told properly that this was how the decisions were being made.

Uber has committed serious infringements. Drivers were deactivated without pardon. From one moment to the next, they no longer had any income through Uber. That's forbidden. A computer should not make decisions on its own that have major consequences for you. These decisions should have been looked at first by a human being.

Monique Verdier, deputy chair of the Autoriteit Persoonsgegevens

The conduct ran from 2018 to 2022 and has since stopped. The company says it strongly disagrees with the decision and with the size of the penalty, that its policies include human review and a route for drivers to dispute a suspension, and that it never automated a permanent deactivation. It has appealed. It also says only 126 drivers in Europe were deactivated on low customer ratings in 2021, and that the penalty is out of proportion to that number.

How the case reached a Dutch desk

It started in France. A hundred and seventy one drivers there took their case to a French civil rights organisation, which put the matter before the French privacy regulator on their behalf. Because the company's European headquarters sit in the Netherlands, the file passed to the Dutch authority under the one stop shop arrangement written into the regulation, and the Dutch regulator investigated with the French one and settled the penalty with the other European supervisors before issuing it.

This is the fourth penalty the same Dutch regulator has imposed on the same company: 600,000 euros in 2018, 10 million in 2023, 290 million in 2024 and now 825 million. The company is contesting the last two of those as well.

European penalties are calculated the same way everywhere and are capped at 4 per cent of worldwide annual turnover. The regulator puts the company's 2025 global turnover at around 44.5 billion euros, so the ceiling was far above what was actually imposed. A decision dated 17 August put the sum at second only to the 1.2 billion euros imposed in Ireland in 2023, which is also under appeal.

Why this one matters outside the taxi trade

The rule engaged here is the prohibition on decisions taken solely by machine where they carry legal or similarly serious effects for a person. The regulation requires meaningful human involvement and a route to challenge the outcome. What the Dutch regulator has priced is not the software itself but the absence of a human being between the software and the consequence.

That reaches a long way past ride hailing. Construction and engineering businesses are putting machine scoring into subcontractor prequalification, site access, competence checking, fatigue monitoring and productivity measurement, and much of it is sold on the promise that nobody has to review the output. Where the output removes somebody's ability to earn, that promise is the liability. A dispute route that exists on paper is not the test either; the Dutch finding is that the company had review processes and still fell foul of the rule because the deactivation itself happened before any human saw it.

The practical question for anyone buying this kind of system is a narrow one. At the moment the software produces an outcome that costs a person work, who signs it off, what are they given to look at, and can they actually overturn it. If the answer is that the system does it and a human hears about it afterwards, the arrangement is the one that has just been fined.