Five complaints accuse Scale AI of tracking the people who train Meta's and OpenAI's AI

Write questions, rank a chatbot's answers, repeat. More than 900,000 freelance workers do this job on Outlier, Scale AI's platform. This is how large AI models learn to answer. Meta, OpenAI, xAI, Google and Anthropic are among Scale AI's customers, according to noyb. Meta owns 49% of the company, Reuters notes.

On October 27, the Austrian nonprofit noyb, founded by Max Schrems, filed five complaints against Scale AI under the EU General Data Protection Regulation (GDPR): one in France, one in Italy, one in the Netherlands and two in Austria. Three partners are supporting the filing: the Paris-based research collective DiPLab, the French nonprofit Intérêt à Agir and Reversing.Works. Scale AI had not responded to Reuters at the time of publication.

What would the platform be monitoring?

According to noyb, while a worker is using Outlier, trackers from TikTok, LinkedIn, Facebook, the survey tool Sprig and Datadog are installed on their device. Every few seconds, they would send these third parties the IP address, contact details, unique identifiers and the pages viewed. This data can be used for ad targeting and profiling. Reuters, which saw the complaints, describes up to 19 trackers. They would record keystrokes and mouse movements. According to DiPLab, this data would make it possible to replay an entire work session.

“Why should LinkedIn or Facebook know what tasks you are working on?” asks Lisa Steinfeld Leiser of noyb. The nonprofit argues that Scale AI has no legal basis to place these trackers without consent.

Their face filmed, over and over

The second complaint is about identity verification. According to noyb, Outlier repeatedly asks workers for a video of their face and an ID document, through the provider Persona. Scale AI cites fraud prevention and skills verification. noyb considers these checks excessive under the data minimization principle, especially since they involve biometric data. The nonprofit points out that in 2025, a breach at Persona exposed about 70,000 photos of ID documents.

The third complaint is about silence. According to Reuters, four workers requested a copy of their data between December 2025 and May 2026. None of them reportedly received a reply, even though the GDPR requires a response within one month in principle.

The invisible labor behind AI

Beyond the trackers, DiPLab describes, based on interviews with workers, a bonus system of $100 to $200 for 200 tasks completed within 24 hours. “You work when the platform wants you to and when tasks are available,” sums up Thomas Le Bonniec, who studied the platform for two years for DiPLab. Le Bonniec points out that the Italian data protection authority fined Glovo in 2021 over similar gamification. noyb is calling for a full investigation and a dissuasive fine. The nonprofit also wants authorities to examine how Outlier distributes tasks.

For European companies that buy training data or have their own data annotated, the case is a reminder of something obvious: annotators' working conditions also fall under the GDPR. It remains to be seen whether the authorities, including the CNIL (the French data protection authority) for the French complaint, will open an investigation, and what Scale AI will say in response.