Meta, Alphabet and xAI stay silent on Scale AI worker surveillance
OpenAI alone has answered for the human-data supply behind its models; Meta, Alphabet and xAI were logged silent today. If regulators side with noyb, the question follows the training data into the buyers' weights, and Meta owns nearly half of the supplier.
Richard Tangthree-minute read
Three of the four companies whose models run on Scale AI's$29B — Scale AI, private, latest valuation $29B training data have chosen silence. The Business and Human Rights Centre, an NGO tracker, asked Meta−0.60% — Meta, down 0.60 percent today, Alphabet+1.10% — Alphabet, up 1.10 percent today, OpenAI$1.18T — OpenAI, private, latest valuation $1.18T and xAI$200B — xAI, private, latest valuation $200B to account for their supplier after noyb, the Austria-based privacy group, filed five complaints against Scale AI under Europe's General Data Protection Regulation in late October; OpenAI replied on November 11, and today the Centre logged the other three as non-responsive 124. We covered the complaints the day they landed; the answer sheet is the news 124.
That silence looks safe and is not. The complaints name a private supplier, but the question they raise travels with the data, into models the buyers have already shipped.
You can swap a supplier in a quarter; you cannot un-train a model.
More than 900,000 freelancers sit inside the weights
Frontier models learn from human feedback, and the humans here are more than 900,000 freelancers on Outlier, Scale's data-work platform, ranking answers and writing prompts for Meta, OpenAI, Alphabet's Google and xAI 2. Working from DiPLab's report *Surveillance at Scale*, noyb alleges that third-party trackers from TikTok, LinkedIn and Facebook fired on workers' devices every few seconds, capturing enough detail to replay a work session as video, and that workers faced repeated face-video identity checks through the verification firm Persona 267.
Four workers who asked Outlier for copies of their data between December and May got no answer at all, noyb says 2. Scale denies acting illegally: its monitoring is disclosed and consented to, screen recording is opt-in and manually activated, and it neither sells worker data nor trains customers' models on it 3. No regulator has ruled; it is allegation against denial 6.
Meta cannot walk away from this one
Meta paid roughly $14 billion last year for a non-voting stake of just under half of Scale 5, which makes its silence the strangest of the three: the owner declining to vouch for its own asset. Alphabet and xAI could move their data work elsewhere; Meta stays invested either way. And noyb's threats carry weight, because its past complaints produced a €1.2 billion fine against Meta itself, and another against TikTok 4.
Silence is the expensive answer
The bottleneck is not the technology; it is supplier standards, and only buyers can enforce those at scale. An audit and a consent warranty in the next data contract cost a rounding error beside a European enforcement process, and answering a letter costs nothing.
I am firmly pro-AI, and that is exactly why this should be cleaned up now. The workers training these models are the low-paid end of the boom, and if their data was gathered under contested consent, the defect sits in everyone's weights, the silent three's included. OpenAI answered a letter. Meta, Alphabet and xAI should say where they stand while a statement is still all it costs.
Get Policy & Labor by email, free
The rules governments write for AI, and what AI does to jobs. In our daily email, every weekday.






Reader comments
0 comments