Kola Oba, 48, was taking a break in Tottenham, north London, sitting beside another Uber driver, when the same job came up on both their screens at the same moment. Oba was offered £23. The driver next to him was offered £27. He described it to the Guardian's Robert Booth, in reporting published on 2 September 2026.

The four-pound gap matters less than the accident that let him see it. Oba and his colleague suspected something specific: that Oba had taken several cheap jobs recently, and the system had learned he would accept less. "It's scary – they have all my information and they are using it against my own wellbeing," he told Booth. "It defines how much I earn, how long I have to work, my time with my family, my resting time." He calls the algorithm "soulless."

Drivers describe a second pattern to Booth, more specific than a price gap: that after a long trip out, the offer to come back is lower — the system, they believe, pricing in the fact that nobody wants to drive home empty.

On 2 September 2026, that suspicion became a lawsuit. The claimant is Stichting WIE International, a Dutch foundation set up for the purpose; the campaign behind it is Worker Info Exchange, whose founder, James Farrar, secured the UK supreme court ruling that Uber drivers should have worker rights. The claim was filed at the Rechtbank Amsterdam — Amsterdam's district court — alleging that Uber's AI-powered pay-setting system personalises the fare offered to each driver, in breach of European data-protection law, and that doing so has pushed earnings down. It also alleges Uber unlawfully used driver data to train its AI models, and seeks damages alongside an injunction to halt the conduct. The Guardian put the number of drivers covered at roughly 241,000 across the EU and UK; the European Trade Union Confederation called it the first collective legal move of its kind. Drivers say the claim could run into billions of dollars. It is not being fought on the drivers' own money: the claim's sign-up page names Innsworth Capital as its litigation funder, entitled to a success fee of up to 25% of a driver's compensation, or 22.5% for members of a union.

Uber's answer, given to the Guardian, arrived early and without hedging. "While we haven't seen the claim yet, we categorically reject the allegations," a spokesperson said, adding that the app uses only real-time trip information — journey, duration, destination — to calculate fares, that drivers see their earnings and where a trip is going before deciding whether to accept it, and that the share Uber keeps from every fare "has remained relatively flat." What dynamic pricing does, the company says, is raise pay on the trips drivers do not want, lifting what a driver can earn rather than pushing it down.

It also denied the precise thing Oba suspects. Uber told the Guardian it does not adjust the price offered for a trip based on an individual driver's behaviour, and that a history of accepting or rejecting trips is not used to personalise pay offers. More than a year earlier, answering separate research into the share it keeps from fares, the company had put the same position in its own words: "Upfront prices are not personalised – our pricing algorithms do not use information about an individual rider or driver's personal characteristics. Suggestions that our systems manipulate pricing unfairly or discriminate are simply false and not supported by evidence."

Everything that follows is an argument about a practice the company says it does not engage in. The claim is untested. No court, in the Netherlands or anywhere else, has found that Uber prices a job by predicting what one particular driver will accept — only that roughly a quarter of a million people, on one count, believe it does, and have now asked a Dutch court to make the company show its work.

Uber has previously attributed discrepancies like Oba's to other features of its system: GPS, surge pricing, promotions, testing. Those four are not the same kind of answer. Two men on one bench are metres apart and inside the same surge zone at the same second, so neither location nor demand has much room to make four pounds. The other two have all the room they need. Individual bonuses run to several pounds and are aimed at particular drivers, though they usually push one number up rather than another's down. And price testing does exactly what Oba saw: same trip, same instant, two offers. That last answer acquits the algorithm of the specific charge, since a randomised test is not a prediction about you. It also concedes the part drivers find hardest to unhear — that what you are offered for a job can already depend on which driver you are. One gap, seen once by accident, cannot tell those two apart.

Uber has never hidden that what a ride costs, and what a driver earns for it, moves with the trip: distance, duration, destination, time of day, local demand. That is trip-based dynamic pricing — pay that varies by journey, not by who is driving it — and Uber has explained it for years as the logic of its marketplace. WIE International's claim does not dispute it. The allegation is that the same machinery has been turned on one named person: that what a driver is offered varies with a prediction about him — his history on the app, what he has accepted before, what the system has learned he will accept again. An automated decision about an identified individual is a different legal object from a market-wide schedule, however finely cut. A price that moves with the cost or the demand of a job is a market. A price that moves with what this particular driver has been observed to take is a judgment about a man.

Uber's own executives have described building that kind of individualised targeting, for trips at least. On the company's fourth-quarter 2023 earnings call, on 7 February 2024, chief executive Dara Khosrowshahi told investors: "So I think that what we can do better is actually targeting of different trips to different drivers based on their preferences or based on behavioral patterns that they are showing us." In the same answer he described upfront fares moving "from just flat time and distance to now kind of point estimates for every single trip based on the driver" — an advantage, he said, that would "accrue to us." Read strictly, he is describing allocation, which trip goes to which driver, and a shift from flat rates to a price computed trip by trip. Whether "based on the driver" reaches price as well, the transcript does not say, and Uber's denials never address the earnings call. The claim alleges that the targeting has since spread from matching to money: from choosing which ride to offer a driver to choosing what to offer him for it. The campaign built its public case on that transcript — a document Uber published to its own shareholders.

The financial stakes, as the claimants tell them, are not small. Drivers told the Guardian that Uber has operated dynamic pay-setting in the UK since 2023, pushing their annual incomes down by about £5,000. That is the claimants' own figure, not an independent audit. It does not say whether the fall is in takings or in take-home after fuel and fees, and the sample is not neutral: the people supplying the figure are the people who joined a claim because they believed they had lost money. Nobody outside the claim can check it yet.

Whether an algorithm could do what the claim alleges is a narrower question, and a 2025 paper in the Journal of the European Economic Association answers only that one. Kuan-Ming Chen, Ning Ding, John List and Magne Mogstad took advantage of a large natural field experiment at Uber — the variation introduced into drivers' ordinary working day rather than a laboratory. Combined with high-frequency data on what drivers then did, it let them recover reservation wages, the least a driver will accept, and show them varying between people and over time within the same person. Such quantities are estimable, with error but with real precision, from data platforms already hold. The paper does not describe what Uber's own pay algorithm does, a line it declines to cross. And a randomised trial that varies what a worker is offered, inside his ordinary working day, is structurally the thing Uber calls testing.

Uber's pay mechanics have drawn outside scrutiny before. In June 2025 the Guardian reported University of Oxford research, drawn from 1.5m UK trips, finding that Uber's median take rate per UK driver — its own share of the fare — had "increased from 25% to 29%, and on some trips … is over 50%." Uber said it did not "recognise the figures," and, asked again this September, that the study relied on incomplete and selective data. That dispute is about the size of the company's cut rather than personalised offers, and forms no part of this claim.

What Oba stumbled into on his break was plainer than an algorithm. It was a comparison, alive for one accidental second because a colleague's screen happened to be visible. A wage does not stop being a wage because only one person is paid it. But nearly every protection built around wages does stop working. Equal-pay claims work by holding two contracts side by side. Collective bargaining works because a union can price many workers with one figure. Each is a comparison engine, and each stalls when there is no second number to set beside the first. A statutory floor survives all of this, because it measures a payslip against a published number rather than against another worker's. Uber says its drivers are guaranteed at least the national living wage, and nothing in this claim disputes it. But a floor catches only the worst outcome. Everything between the floor and what the job is actually worth has always been policed by comparison, and it is that whole range the claim says is going dark. A driver who cannot see what anyone else was offered cannot easily tell what his hour is worth elsewhere, and a worker who cannot tell is a worker who is cheaper to buy. That argument is this piece's construction rather than a point any lawyer in the case has made in public: the wrong alleged here is the loss of the rate itself, not a payment beneath it. Take away the twin and there is nothing left to underpay against — only a number nobody else received.

That is also why the case runs through data-protection law rather than labour law. Employment statutes were not written for an offer with no comparable twin; there is no established wage doctrine for "the system set my pay just for me." Data-protection law does have a doctrine for a decision made about one identified person by automated means — the GDPR calls it automated decision-making, including profiling — and that sits far closer to what is being alleged. The claim is framed in exactly those terms, alleging that Uber "unlawfully used automated decision-making, including profiling" — a privacy claim carrying a wage argument inside it, not the other way round. Anton Ekker, the Dutch lawyer leading the case, told the Guardian on 2 September 2026 that "a computer algorithm should not independently make decisions that strip individuals of their livelihood," arguing that platforms should be "held accountable for the large-scale exploitation of vulnerabilities of European citizens."

A Dutch court has looked inside this system once before, on a much smaller question. On 4 April 2023 the Amsterdam Court of Appeal — a different bench from the district court now hearing the new claim — ruled on a request by Uber drivers for access to their own data under Article 15 of the GDPR. It ordered Uber, on pain of a penalty payment, to hand over several categories of that data and to say whether automated decision-making including profiling had been applied — and where it had, to explain the logic involved and its consequences. The ruling is narrower than it is sometimes given credit for. The court compelled an answer, not a verdict, and the manipulation of pay was never the question before it. What it shows is a court willing to make Uber open its systems to inspection: procedural ground already gained rather than a merits finding the new claim can borrow.

The law is about to close part of this gap, though not yet. Directive (EU) 2024/2831, the EU's platform-work directive, names the mechanism in its eighth recital, among the managerial functions algorithms increasingly perform in place of a human manager: "allocating tasks, pricing individual assignments, determining working schedules… evaluating the work performed." A later recital extends that to "a decision, the lack of a decision or a set of decisions" made by such systems, and singles out a worker's "potential earnings" among the effects it treats most seriously. The directive will eventually require platforms to disclose the main parameters an automated system weighs, and how a worker's own data and behaviour move the result. That duty binds nobody until December 2026. For now, a Dutch judge is being asked to find, under today's data-protection law, a version of a wrong Europe's legislature has already named and not yet required any platform to explain.

Underneath the law sits a harder question the case never quite asks a court to answer: whether any of this is bad economics.

Here it is — not anything Uber has said in public, but the argument this piece constructs on the company's behalf, at full strength. Grant that a pay system really could learn what each driver would accept for a given trip, and price to it. Economists have a name for it: first-degree price discrimination. A single posted rate always turns some workers away, because anyone whose walk-away price sits above it declines. Most of the time that costs nothing; the job routes to the next driver. But in the thin moments — four in the morning, the outer suburbs, a weather spike — there is no next driver, and the trip goes unfilled. Price to each worker's own floor instead, and the turned-away worker disappears: the job gets done, at a wage that worker was willing to take all along. Trips nobody would have taken at the flat rate now get taken, a real gain shared with the riders who otherwise got no car. The rest is a transfer: the money that used to go to drivers who would have driven for less than the flat rate anyway, moving to whichever side now prices closer to their floor. On the standard test, total surplus rises. And every one of these offers is accept-or-decline with the number shown first, so no driver ends up worse off than if the trip had never appeared at all. The dispute is about how a profitable trade gets split. Nobody was forced into anything.

The argument runs out there: efficiency was never the drivers' complaint. And little of that efficiency is actually on offer here. The unfilled trip is a genuine problem at four in the morning on the edge of town. It is close to a non-problem at six on a Friday evening, when there are more drivers than jobs and a refusal costs the system a few seconds. Take away the unfilled trips and almost nothing remains but the transfer. Nor does accept-or-decline settle the question of consent, because a refusal is only free if the offers you see tomorrow are independent of what you turn down today — and under a system that is learning about you, they are not. The textbook case assumes one throw of the dice. This one runs every day, on a driver who knows it is watching. Once taking a cheap job today means being offered less tomorrow, the rational move is to refuse work you would otherwise have wanted, purely to keep your own price up — and that is waste, of a kind a flat rate never produced. Mohammed Shirwa, a 41-year-old driver in Rotterdam, put it to Booth more precisely than any paper has: "All the time the algorithm is learning about you and what you are willing to accept. So the prices go low but you are stuck. It knows you need the job."

What a single posted rate paid out, without anyone designing it to, was an overpayment: every driver whose own floor sat below the posted number got the difference, free. Economists call that an informational rent, and the name is exact — the driver was paid it not for skill or speed but because the company could not see what he would have settled for. It is the price of not being known. Individualise the offer, and that rent gets a new owner. Whose, exactly, this piece cannot settle. If competition for riders is fierce, the money leaves again as cheaper fares, which is close to what Uber implies when it says its share of each fare has stayed flat. If drivers can switch apps easily, a rival's offer caps what any personalised number can take. And the drivers who were turned away under the flat rate are better off, not worse. What the claimants allege is the case where none of that holds: the fares did not fall, the switching is harder than it looks, and the money simply stayed. Wage law exists to stop exactly this kind of quiet transfer, not to chase efficiency, and the claimants' position, stated plainly, is that a system which prices each person alone is the transfer, working as designed. Veena Dubal, a professor of law at the University of California, Irvine, made the underlying move visible in the Columbia Law Review in 2023, describing it as "the importation of price discrimination from the consumer context to the labor context." Translate that from theory into a wage claim, and the shape doesn't change: a job that pays each person a different, privately computed number does not simply pay some of them less. It removes the number they would need even to notice.

None of this has yet been tested by a judge. As of 10 September 2026, a week after the claim was filed, Uber had not filed a defence, no hearing date had been set, and the company maintained it had yet to review the filing at all. Amsterdam moves on its own schedule.

What Uber files there will be the first real test of whether a comparable rate survives once the price is different for everyone who is offered it. It is the test Kola Oba ran by accident, on a bench in Tottenham, on his break. Directive (EU) 2024/2831 already names the practice; its disclosure duty binds no platform until December 2026, three months after a Dutch court got the question first.

Sources

  • Robert Booth, "Uber drivers launch European class action over 'soulless' and 'scary' AI algorithm," The Guardian, 2 September 2026. Source of the Oba and Shirwa accounts and quotations, the claim's scope and valuation figures, Uber's September 2026 responses, and the £5,000 earnings estimate.
  • "Second study finds Uber used opaque algorithm to dramatically boost profits," The Guardian, 25 June 2025. Source of the University of Oxford take-rate findings reported the previous week, and of Uber's June 2025 statement that upfront prices are not personalised.
  • Uber Technologies, Inc., Q4 2023 earnings call, corrected transcript, 7 February 2024 (FactSet CallStreet). Source of the Khosrowshahi quotations.
  • Stichting WIE International, claim sign-up page, including the litigation-funding terms.
  • Gerechtshof Amsterdam (Amsterdam Court of Appeal), ECLI:NL:GHAMS:2023:796, case 200.295.747/01, judgment of 4 April 2023, published in Dutch on rechtspraak.nl.
  • Directive (EU) 2024/2831 on improving working conditions in platform work, cited from the signed Council text PE-CONS 89/1/24 REV 1. The directive's disclosure duty applies from December 2026; no primary text consulted for this article carries a day within that month.
  • Kuan-Ming Chen, Ning Ding, John A. List and Magne Mogstad, "Reservation Wages and Workers' Valuation of Job Flexibility: Evidence from a Natural Field Experiment," Journal of the European Economic Association, published 10 June 2025, doi:10.1093/jeea/jvaf022.
  • Veena Dubal, "On Algorithmic Wage Discrimination," Columbia Law Review, vol. 123, no. 7 (2023).