Companies Tap Worker Data to Train Autonomous AI, Stirring Privacy Concerns

Developers of autonomous AI and humanoid robots increasingly need real-world workflow data that text and simulations cannot fully reproduce, turning employees into sources of training material. Indian garment maker Pearl Global Industries has equipped workers with Egolab AI headsets that record egocentric video of sewing, folding and handling fabric. The push dovetails with lights-out manufacturing: China installed 295,000 industrial robots in 2024, or 54% of the global total, according to the International Federation of Robotics.

Meta rolled out its Model Capability Initiative in April 2026 to capture mouse movements, clicks, keystrokes and screen content from U.S.-based employees, aiming to train AI agents to navigate software without supervision. The company said on June 22 it would pause the program while investigating data-security concerns after sensitive records were found accessible to all staff. More than 1,600 employees signed a petition opposing the tool. Meta expects 2026 capital spending of as much as $145 billion, much of it tied to AI.