According to Wired, more than 200 contractors who supported Google’s AI projects were abruptly dumped in August. There was no prior warning, even though these workers were the ones training, rating, and correcting the output that makes Google’s models look clever.

The so-called “super raters”, hired through outsourcing firm GlobalLogic, were on $28 to $32 (€26 to €30) an hour. Others doing near-identical work were paid $18 to $22 (€17 to €20). Staff complained of poor job security, a lack of benefits, and the kind of pay disparity that usually sparks union talk.

Workers started sharing frustrations online, only to find themselves slapped with warnings. Some claim they were then pushed out after being assigned “unrealistic tasks” designed to pile on stress. When one worker asked why they were being laid off, the only explanation was a vague line about “ramping down a project.”

Google is keeping quiet about the whole thing and insists responsibility for workplace conditions sits with its contractors and subcontractors.

What these cuts show is that even as AI systems automate more tasks, they still need armies of low-paid humans to clean up the mess. And those humans, at least at Google, seem to be expendible.