On 27 October 2026 the Vienna-based privacy group noyb filed five GDPR complaints against Scale AI, the US company behind Outlier, a data annotation and AI-training platform. The complaints were filed by noyb in collaboration with the French organisation Intérêt à Agir and the research groups DiPLab and Reversing.Works under Article 80(1) of the GDPR, which lets organisations bring claims on a person’s behalf. The filings belong to AI-WORLD, an interdisciplinary initiative on working conditions across AI value chains led by Intérêt à Agir with DiPLab. Each filing names one identified worker who lives and works in the country concerned.
The five split across four authorities: two went to Austria’s Datenschutzbehörde (DSB) in Vienna, and one each to the CNIL in France, the Garante per la protezione dei dati personali in Italy and the Autoriteit Persoonsgegevens (AP) in the Netherlands. That detail comes from the complaint documents themselves, whose addressee blocks noyb has published. None of the four has confirmed receipt, opened a procedure or set a timetable. The complaints ask each authority to investigate, issue a declaratory decision, order Outlier to answer access requests and impose a fine. Those are requests, and no data protection authority (DPA) has ruled on them.
Trackers, biometric checks and unanswered access requests
The complaints rest on a technical analysis by DiPLab and Reversing.Works of network traffic captured while complainants used Outlier. That analysis found third-party trackers loading on workers’ devices as they worked, sending IP addresses, contact details, unique identifiers and the pages they visited to outside companies. The noyb release names TikTok, LinkedIn, Facebook, the survey tool Sprig and the analytics company Datadog. The complaint texts list more vendors, and the sets differ between filings. One session-replay tool, Fullstory, is described as recording mouse movements, clicks and keystrokes and sending data every few seconds. Because the finding comes from the complainant coalition’s own captures, it has not been independently replicated.
A second strand concerns identity verification. The complaints say Outlier requires workers to complete repeated biometric checks run by the vendor Persona, including video of their face and an upload of their ID. noyb argues this goes beyond what the task requires, pointing to less intrusive options such as two-factor authentication or a hardware key and to a 2024 Garante decision that found biometric verification of platform workers was not necessary to provide the service. The complaints also cite recital 41 of the EU platform-work Directive, which is not yet transposed and says that where biometric processing does take place, one-to-one verification should be preferred to one-to-many identification.
Third, the complaints say Outlier did not answer requests from workers asking what data it holds on them. Four workers made requests between December 2025 and May 2026, according to reporting on the filings; the complaints say none received an answer.
What the GDPR requires of trackers, biometric data and access requests
Under the GDPR, a company needs a valid legal basis to process personal data. For trackers placed on a device, EU rules separately require consent. Data minimisation is the principle that a business should collect only what it needs for a stated purpose. The right of access lets a person ask a company for a copy of their data, and the company normally has one month to respond. The complaints allege breaches across all three areas. They also cite Article 25(1) and Article 28(1), which cover data protection by design and the choice of processors.
The 900,000 figure is the platform’s own
noyb’s release says Outlier “employs more than 900,000 ‘freelance’ workers”. That number originates with Scale AI’s own marketing. Outlier’s website invites people to “Join 900K+ MAs, PhDs, and college graduates working and earning as AI trainers” and says the company has “helped over 900K experts” gain experience on the platform. It is a cumulative figure across the platform’s life, not an audited count of active workers. The complaints themselves are more cautious, saying only “hundreds of thousands”. By comparison, Fast Company Middle East quoted Outlier’s general manager in November 2024 describing “tens of thousands of freelance contributors” active over roughly the preceding year.
‘Freelance’ status is contested
The complaints describe Outlier’s workers as hired as independent contractors rather than employees, and argue that this treatment is wrong because it leaves them outside social benefits. That is the coalition’s characterisation. Scale AI’s terms do make contributors contractors, and the company says workers choose their tasks, know the pay before accepting and see rates set by skill, project and location. Employment status in platform work is a live legal question across the EU, and this article takes no position on it.
Persona’s October 2025 breach sits behind the processor claim
In October 2025, about 70,000 images of government-issued IDs held by Persona were exposed in a hack. The complaints cite that breach as supporting context for their claim that Scale AI chose a processor that cannot guarantee data minimisation and security. The biometric argument itself does not depend on the breach. No statement from Persona on these complaints was found.
Scale AI denies the allegations
Scale AI rejected the claims on the record. “The report’s claims are unfounded and inaccurate,” a spokesperson told The Register. The company said its practices are disclosed to contributors and require their consent, and that it uses standard tools to verify identities, prevent fraud, protect confidential information and ensure accurate payment. To the Italian outlet IRPI, Scale AI said it does not sell contributors’ personal data to third parties or use it to train clients’ models, and that some projects use a tool recording task time and screen activity, which contributors activate manually after notice. It also said contributors choose tasks and know the pay in advance, with access varying by client need.
Who bears the cost of training data
The named customers show where the platform’s output goes. OpenAI, Meta, Google, xAI and the US government are supported by other sources, including a leaked Scale AI project database described by Inc. and Scale AI’s own blog on its expanded agreement with the Pentagon’s CDAO. Anthropic appears in noyb’s customer list but not in the independently sourced ones, so that claim stays attributed to noyb rather than stated as established.
Scale AI’s platform turns on labour that is geographically dispersed and paid by task. The same systems that check identity, record screen time and prevent fraud are the ones collecting the data at issue, which is why the complaints treat the tracking and the employment terms as one story rather than two.
What happens next is largely out of public view. The GDPR requires a supervisory authority to keep a complainant informed of progress and outcome, but sets no near-term deadline, and noyb’s record in similar cases shows proceedings that run for years. The visible signal will be a confirmation from the CNIL, the Garante, the AP or the Austrian DSB that it has opened a procedure.
