Oct 29, 2026 · 8 min read
noyb Files 5 GDPR Complaints Over Scale AI Worker Tracking
On October 27, 2026, noyb and three partner groups asked data protection authorities in France, Italy, the Netherlands and Austria to fine Scale AI over trackers on its Outlier work platform, repeated Persona face scans and access requests that went unanswered for months.
If you label data for Outlier, a Facebook tracker may know which project you are assigned to and why you have no tasks today. That is one finding in five GDPR complaints noyb filed against Scale AI on October 27, 2026, with Intérêt à Agir, the research group DIPLab and Reversing.Works.
One correction to early coverage. The Record framed the filings as noyb suing Scale and wrote that noyb is "asking courts in the four countries to issue fines." These are not lawsuits. They are complaints to national regulators: the CNIL in France, the Garante in Italy, the Autoriteit Persoonsgegevens in the Netherlands and Austria's Datenschutzbehörde, which received two.
Key Takeaways
- noyb filed five GDPR complaints against Scale AI on October 27, 2026 with the data protection authorities of France, Italy, the Netherlands and Austria, which received two.
- The French complaint lists trackers from 18 third parties on the Outlier platform, including Facebook, LinkedIn, TikTok, Snapchat and the session replay tool Fullstory.
- Articles 5(1)(c), 12(3), 15, 25(1) and 28(1) GDPR are cited in all five complaints, alongside each country's national law requiring consent for trackers.
- Outlier received an access request from each complainant, with the dated ones sent between December 17, 2025 and February 10, 2026, and the complaints say none was answered.
- Scale AI told The Record: "The report's claims are unfounded and inaccurate." It said screen activity is only recorded when workers opt in.
What Is noyb Accusing Scale AI Of?
noyb accuses Scale AI of breaking the GDPR on Outlier in three ways: loading third party trackers without consent or necessity, forcing workers through excessive biometric checks run by Persona, and ignoring requests for their own data.
Outlier is Scale's data annotation platform. According to noyb, more than 900,000 "freelance" workers use it to train AI for customers such as Meta, OpenAI, xAI and Google, and The Record notes that Meta owns nearly half of Scale. Workers are classed as contractors, but the GDPR protects their data whatever their contract says.
What Did the Traffic Analysis Find?
The complainants recorded their browser traffic as HAR files while using Outlier, and the logs show worker data flowing to advertising, analytics and support vendors during ordinary sessions. The redacted French complaint names 18 third parties, from Fullstory and Google Analytics to Quora and Snapchat. One Austrian complaint adds Datadog. The details matter more than the count:
- Facebook: requests carried a worker team ID, a project ID, whether the worker had a task and why not (no task matched their expertise), plus identifiers Meta uses to tie site visits to browsers for ad targeting.
- Fullstory: a session replay tool that records mouse movements, clicks and keystrokes. On Outlier it sent a request every few seconds, and fields such as SetInputValue reveal what is typed into forms. The Record reports noyb's view that the data can generate video of a freelancer's entire workday.
- Forethought and Sprig: a support tool and a survey tool whose trackers carried workers' names and email addresses, and in Sprig's case a phone number, while no survey was running.
"Why should LinkedIn or Facebook know which tasks you're working on?" asked Lisa Steinfeld Leiser, a data protection lawyer at noyb. The pattern echoes Meta's keystroke tracking of its own employees for AI training.
Which GDPR Articles Do the Complaints Cite?
Every complaint cites Articles 5(1)(c), 12(3), 15, 25(1) and 28(1) of the GDPR, plus the national rule implementing Article 5(3) of the ePrivacy Directive:
- Minimisation, Article 5(1)(c): IP address, device ID and even the site visited before Outlier are collected. noyb argues none is needed to run a labelling job.
- Legal basis: Scale relies on legitimate interest under Article 6(1)(f). Citing the CJEU's Meta v Bundeskartellamt ruling, the French complaint argues the minimisation breach invalidates that basis.
- Tracker consent: ePrivacy requires consent, which legitimate interest cannot replace. The complaints invoke Article 82 of France's Loi Informatique et Libertés, Article 122(1) of Italy's privacy code, Article 11.7a of the Dutch Telecommunicatiewet and §165(3) of Austria's Telekommunikationsgesetz 2021.
- Biometrics, Articles 25(1) and 28(1): Scale chose a processor that collects far more than identity matching needs.
- Access, Articles 15 and 12(3): no reply within one month.
One procedural point deserves attention. All five complaints argue Scale has no main establishment in the EU, so under Article 77 each national authority is competent. No lead authority, no one stop shop: a US platform without an EU base can face four regulators moving at four speeds. Each complaint asks for a fine under Articles 58(2)(i) and 83(5)(a) and (b).
Why Is the Persona Biometric Check Under Fire?
noyb argues that making workers record face video and upload ID documents, repeatedly and at irregular intervals, is not necessary to fight account fraud. Approximately 70,000 photos of government IDs stored with Persona were exposed by hackers in October 2025, per noyb, and the French complaint says a February 2026 source code exposure showed Persona can run 269 distinct checks per identity verification.
There is direct precedent. In a November 13, 2024 decision on Foodinho, Glovo's Italian subsidiary, the Garante examined facial recognition checks on delivery riders; TechCrunch reported a €5 million fine and a ban on using riders' biometric data for identity verification. The French complaint cites that decision for both its biometric and tracker claims. Scale told The Record that identity checks exist to combat fraud and comply with European law.
The Platform Work Directive (EU) 2024/2831 must be transposed by 2 December 2026, five weeks after these filings. The French complaint points to its recital 41, which favours one to one verification (face against ID) over one to many searches.
The Access Requests Nobody Answered
Most coverage leads with the trackers, but the strongest claim may be the dullest. Whether a tracker is "strictly necessary" invites argument. Whether a company answered an access request within a month does not.
The complainants wrote to Outlier on dates running from December 17, 2025 (Italy) to February 10, 2026 (Austria), several with follow ups, and none got a reply. Requests to Persona on May 12 and May 25, 2026 drew only an acknowledgement. The Record summarised this as four freelancers asking between December 2025 and May 2026; the filings show five complainants writing to Outlier, with the May dates belonging to Persona.
Each request asked whether automated decision making had been applied, a pointed question on a platform where algorithms assign tasks and pay drops when a timer runs out. The Dutch regulator has already fined Uber €824,990,000 for automated driver deactivations, and the French complaint asks the CNIL to extend its probe to Scale's automated decisions.
What This Means for Your Inbox
The French complaint defines a tracking pixel using the CNIL's own wording: an invisible image whose loading tells a server that a tracked user visited a page or read an email. The Meta, LinkedIn and TikTok tags inside Outlier come from the same ad networks whose pixels sit in marketing emails, and Italy now requires prior consent for email tracking pixels on the same device access logic.
Your email address is also the join key. Forethought and Sprig attached workers' addresses to tracking requests, letting vendors link work activity to everything else tied to that address. And every access request in this case went out by email: the sent message, with its timestamp, is the proof that the one month clock started.
What Should Compliance Teams and Platform Workers Do Now?
Compliance teams should audit their own worker portals the way noyb audited Outlier, starting with a logged in traffic capture:
- Capture a HAR file of your authenticated portal and list every third party domain it calls.
- Remove ad network tags (Meta, LinkedIn, TikTok, Snapchat) from logged in work tools, and put other nonessential trackers behind consent.
- Restrict session replay along the lines of the CNIL's draft session replay recommendation, cited in the complaint: record occasionally and at random, mask or hash inputs, avoid unique identifiers.
- Limit identity vendors by contract under Article 28: one to one matching, no reuse for training.
- Monitor every privacy mailbox named in your policy, including any EU representative's address, against a one month deadline.
Platform workers can make the same request the complainants made:
- Email every privacy address in the platform's policy. The complainants wrote to privacy@outlier.ai and outlier@lionheartsquared.eu, both listed by Outlier.
- Cite Article 15 GDPR and ask for a copy of your data, its recipients including tracker vendors, retention periods, and whether automated decision making is used.
- Send a separate request to the identity vendor, such as Persona.
- Keep the sent email, wait one month, send one reminder, then complain to your national data protection authority.
Looking Ahead
Four regulators now hold near identical files with no shared deadline. The Garante has its Foodinho reasoning ready, and the platform work deadline lands in December. Scale disputes every claim, and no regulator has ruled. Thomas Le Bonniec of DIPLab put the stakes plainly in noyb's announcement: "Platform work seems to be fertile ground for a whole new type of workplace surveillance."