The Amsterdam Court of Appeal ruled Uber deactivated drivers by automated decision under Article 22 GDPR, rejected its 'human in the loop' as a symbolic act, and ordered disclosure under a 4,000-euro-a-day penalty (2023)
On 4 April 2023 the Gerechtshof Amsterdam (Amsterdam Court of Appeal) ruled that Uber's deactivation of drivers' accounts was automated decision-making within the meaning of Article 22 GDPR, held that any human oversight must be meaningful and 'not merely a symbolic act', and ordered Uber to disclose the underlying logic within one month under penalty (dwangsom) of 4,000 euros per day of non-compliance. Every figure is quoted verbatim from the court's own published rulings (rechtspraak.nl) and corroborated by independent press (TechCrunch).
| Metric | Before | After |
|---|---|---|
| On 4 April 2023 the Amsterdam Court of Appeal found Uber's deactivation of drivers' accounts was automated decision-making within the meaning of Article 22 GDPR (Gerechtshof Amsterdam, 2023; TechCrunch, 2023) | ||
| The court ordered Uber to disclose useful information about the underlying logic of the processing within one month under penalty of 4,000 euros per day of non-compliance (Gerechtshof Amsterdam, 2023) | ||
| The court held that human oversight of an automated decision must be meaningful and 'not merely a symbolic act', rejecting Uber's reliance on human review (Gerechtshof Amsterdam, 2023; TechCrunch, 2023) | ||
| The automated systems examined included assigning rides, calculating prices, rating drivers, fraud-probability scores and account deactivation; a linked ruling ordered Ola Cabs to explain its fraud-probability scores (TechCrunch, 2023) | ||
Verification status: PENDING — every figure is quoted verbatim from the Amsterdam Court of Appeal’s own published rulings and independent press (TechCrunch), and re-checked by the checker; graduated to pending, awaiting the owner’s final green sign-off.
The problem
The dispute was not about one firing but about the machine that manages the driver. A group of Uber drivers, supported by Worker Info Exchange and the App Drivers & Couriers Union, went to the Dutch courts to force disclosure of how Uber’s algorithms decided their working lives after their accounts were switched off. The Amsterdam Court of Appeal framed the case as being about “informatie over het bestaan van geautomatiseerde besluitvorming in de zin van artikel 22 AVG) nadat hun accounts door Uber waren gedeactiveerd” (information about the existence of automated decision-making within the meaning of Article 22 GDPR, after their accounts had been deactivated by Uber) (source). Independent technology press listed the systems in question as “assigning rides; calculating prices; rating drivers; calculating ‘fraud probability scores’; and deactivating drivers’ accounts in response to suspicions of fraud” (source).
What was built
The systems under scrutiny were Uber’s algorithmic driver-management tools, and the legal question was whether the account deactivations they produced were, in law, decisions taken by a machine. Article 22 GDPR, as the court quoted it, gives a person “het recht niet te worden onderworpen aan een uitsluitend op geautomatiseerde verwerking, waaronder profilering, gebaseerd besluit waaraan voor hem rechtsgevolgen zijn verbonden of dat hem anderszins in aanmerkelijke mate treft” (the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects or similarly significantly affects them) (source). Uber’s defence was that a person, not only the algorithm, made the call. The court set the bar that defence had to clear: the controller “moet ervoor zorgen dat al het toezicht op de besluitvorming zinvol is, en niet slechts een symbolische handeling vormt. Deze tussenkomst moet worden uitgevoerd door iemand die bevoegd en bekwaam is om het besluit te veranderen” (must ensure that all oversight of the decision-making is meaningful and not merely a symbolic act; this intervention must be carried out by someone competent and authorised to change the decision) (source). TechCrunch reported that Uber “could not demonstrate the staff involved had been able to exercise a meaningful check on the automated decision that led to drivers being fired” (source).
The outcome
The court ordered disclosure, and it attached a price to delay. In its operative decision it ordered Uber, “op straffe van een dwangsom van € 4.000,00 per dag of deel daarvan dat Uber in gebreke blijft, binnen een maand na betekening van de beschikking” (under penalty of 4,000 euros for each day or part of a day that Uber remains in default, within one month of service of the decision), to grant the drivers’ Article 15(1)(h) GDPR requests and provide “nuttige informatie over de onderliggende logica, alsmede het belang en de verwachte gevolgen van die verwerking” (useful information about the underlying logic, as well as the significance and the expected consequences of that processing) (source). The parallel Uber ruling imposed the same daily penalty, awarding the dwangsom “tot € 4.000,00 per dag of dagdeel, in lijn met de dwangsom zoals die wordt toegewezen in de hiervoor reeds genoemde (door andere chauffeurs tegen Uber aangespannen) zaak” (up to 4,000 euros per day or part-day, in line with the penalty awarded in the aforementioned case brought by other drivers against Uber) (source).
The reach went beyond Uber. TechCrunch reported that the same court also ordered Ola to disclose information about the “fraud probability scores” it used in automated decisions about work and pay, and that the platforms could not withhold the information as trade secrets, a claim the court found “entirely disproportionate relative to the negative effect of unexplained automated dismissal and disciplining of workers” (source).
Weakest load-bearing source. The Ola limb of the story and the “assigning rides / fraud probability scores” enumeration rest on a single independent secondary source, TechCrunch, rather than on the Ola primary ruling, which was not fetched this session; they are presented as reported, and nothing in the core outcome depends on them. Every finding that carries the story, the Article 22 characterisation of the deactivations, the one-month disclosure order, the 4,000-euro-per-day dwangsom and the “symbolische handeling” standard, comes from the Amsterdam Court of Appeal’s own published rulings (the adjudicating body, independent of the subject Uber) and is corroborated by TechCrunch. No confirmation was sought from Uber; the rulings are used only as already-public official documents.
How this was verified. Method: every load-bearing figure is quoted verbatim from a source fetched and grepped this session and saved to sources/. The Article 22 GDPR finding on the account deactivations, the order to disclose “nuttige informatie over de onderliggende logica”, the one-month deadline, the 4,000-euro-per-day dwangsom and the standard that human oversight must be “zinvol … en niet slechts een symbolische handeling” are quoted from the Amsterdam Court of Appeal’s published rulings ECLI:NL:GHAMS:2023:793 and ECLI:NL:GHAMS:2023:796, both dated 4 April 2023 (Tier 1, the adjudicating body, independent of the subject; retrieved from the official rechtspraak.nl content API and saved to sources/rechtspraak-ghams-2023-793.xml and sources/rechtspraak-ghams-2023-796.xml). They are independently corroborated by TechCrunch, “Drivers in Europe net big data rights win against Uber and Ola” (5 April 2023, Tier 2; Wayback 20251121100017, saved to sources/techcrunch-2023-04-05.html), which also carries the Ola fraud-probability-score and trade-secret findings. No confirmation was sought from Uber. Checked 2026-09-03 (checking round 1).
Related case files
- The EEOC’s consent decree over iTutorGroup’s hiring algorithm, where an automated rule making a real employment decision was held unlawful — the closest analogue: an authority forcing accountability for an algorithm that decided who could work, here on age rather than fraud scores.
- Cigna’s PXDX algorithm, which a US court let a claim proceed against for automated decisions at scale — the same question of whether a decision taken in seconds by software, with only nominal human review, can still be laid at a company’s door.
- The Dutch DPA’s 30.5-million-euro fine of Clearview AI for an unlawful facial-recognition database — another Dutch enforcement of the GDPR against an opaque algorithmic system, useful for comparing a regulator’s fine with a court’s disclosure order.
- Tesla’s NHTSA-forced recall of 2,031,220 vehicles over Autopilot — another transport-sector automated system that an authority compelled to change, told through a safety recall rather than a data-rights ruling.
Sources
- Gerechtshof Amsterdam (Amsterdam Court of Appeal) · ECLI:NL:GHAMS:2023:793, zaaknummer 200.295.742/01 · 4 April 2023 · https://data.rechtspraak.nl/uitspraken/content?id=ECLI:NL:GHAMS:2023:793 — Tier 1 (primary record from the adjudicating body, independent of the subject Uber; the Article 22 GDPR finding on the account deactivations, the order to disclose “nuttige informatie over de onderliggende logica”, the one-month deadline, the 4,000-euro-per-day dwangsom, and the “symbolische handeling” standard for human oversight; retrieved from the official rechtspraak.nl content API, saved to sources/rechtspraak-ghams-2023-793.xml).
- Gerechtshof Amsterdam (Amsterdam Court of Appeal) · ECLI:NL:GHAMS:2023:796 · 4 April 2023 · https://data.rechtspraak.nl/uitspraken/content?id=ECLI:NL:GHAMS:2023:796 — Tier 1 (primary record from the adjudicating body; the Article 22 GDPR text as quoted by the court and the parallel 4,000-euro-per-day dwangsom “in lijn met” the linked Uber case; saved to sources/rechtspraak-ghams-2023-796.xml).
- TechCrunch · Drivers in Europe net big data rights win against Uber and Ola (Natasha Lomas) · 5 April 2023 · https://techcrunch.com/2023/04/05/uber-ola-gdpr-worker-data-access-rights-appeal/ — Tier 2 (independent technology journalism; corroborates the Article 22 finding, the rejection of Uber’s human-review defence, and carries the Ola fraud-probability-score and trade-secret findings; Wayback 20251121100017, saved to sources/techcrunch-2023-04-05.html).
Uber's algorithmic driver-management systems: automated account deactivation, batched ride matching, upfront pricing, and driver ratings; Ola Cabs' fraud-probability scoring (linked case)
- Status
- pending
- Method
- Every load-bearing figure is quoted verbatim from a source fetched and saved to sources/ this session. The Article 22 finding, the one-month disclosure order, the 4,000-euro-per-day dwangsom and the 'symbolische handeling' (symbolic act) standard are quoted from the Amsterdam Court of Appeal's own published rulings ECLI:NL:GHAMS:2023:793 and ECLI:NL:GHAMS:2023:796 (rechtspraak.nl, Tier 1, the adjudicating body, independent of the subject Uber; saved to sources/rechtspraak-ghams-2023-793.xml and sources/rechtspraak-ghams-2023-796.xml). They are independently corroborated by TechCrunch, 5 April 2023 (Tier 2; Wayback 20251121100017, saved to sources/techcrunch-2023-04-05.html). No confirmation was sought from Uber; the rulings are used as already-public official documents.
- Provider
- Gerechtshof Amsterdam (Amsterdam Court of Appeal), the adjudicating body; the subject is Uber B.V. and its algorithmic driver-management systems
- Client
- Uber B.V. (subject; the Amsterdam Court of Appeal is the adjudicating body, the drivers represented by Worker Info Exchange and the App Drivers & Couriers Union) · transport
- Disclosure
- named
What did the Amsterdam Court of Appeal decide about Uber's drivers on 4 April 2023?
It ruled that Uber's deactivation of the drivers' accounts was based on automated decision-making within the meaning of Article 22 GDPR, and ordered Uber to give the drivers useful information about the underlying logic, significance and expected consequences of that processing within one month of service, under penalty of 4,000 euros for each day of non-compliance.
Why did the court reject Uber's 'human in the loop' defence?
The court held that a controller must ensure that all oversight of the decision-making is meaningful and not merely a symbolic act, carried out by someone competent and authorised to change the decision. On the facts it found Uber's human review did not meet that bar, so the deactivations counted as solely automated decisions.
Was Ola Cabs involved?
Yes. In a linked ruling the same court addressed Ola Cabs' automated systems, including 'fraud probability scores', and ordered disclosure of the logic behind them. The cases were brought by drivers supported by Worker Info Exchange and the App Drivers & Couriers Union.