Uber headhunted PhDs to join 'Project Sandbox.' After a month, it said that their AI training contracts were over. (www.businessinsider.com)

Uber has abruptly cut short an AI-training engagement called "Project Sandbox," telling about a dozen contractors — including several with PhDs — that their services are no longer needed because the client “communicated a change in their internal priorities.” Uber had recruited these workers this fall to perform tasks for Google under its AI Solutions arm, onboarding them through a staffing agency and placing them in direct contact with Google employees. Work included photo/video annotation and evaluation of AI-generated answers; pay advertised ranged from about $55 to $110/hour if contractors could secure consistent 40-hour weeks. Contractors say they were promised at least three months of work but were told the assignment would end immediately, with some still awaiting their first paycheck (which Uber said could take up to seven weeks). For the AI/ML community this episode underscores two realities: the heavy reliance on contingent, often highly skilled labor for data-labeling and model evaluation work, and the operational fragility of outsourced pipelines when client priorities shift. The incident highlights ethical and practical problems — variable pay tied to unpredictable hours, short notice cancellations, and delayed payments — that can disrupt training data supply, auditability, and workforce stability. As platforms like Uber expand into “digital tasks” for AI, researchers and companies should plan for greater contractual transparency and redundancy in labeling pipelines to avoid sudden data bottlenecks.