Meta Halts Worker-Surveillance AI Program After Internal Data Exposure

Meta has suspended its Model Compatibility Initiative (MCI), a controversial program that captured employees’ keystrokes, mouse movements, clicks, and screen contents to train AI systems on how humans operate software. The pause came after an internal security notice revealed that databases holding MCI-collected data had been left accessible to anyone inside the company. Meta says it found no evidence that employees improperly accessed the data, but is investigating before deciding whether to restart the program.

According to VP of AI research Stephane Kasriel, the exposure was first flagged on June 18 and patched within four hours, but the initial fix failed to hold and access had to be locked down again. Rolled out to US staff in April, MCI originally offered no opt-out; that changed only partially after employees petitioned over privacy, security, and personal-liberty concerns. Leadership had repeatedly defended the collection as essential, arguing workers were the ideal models for teaching AI to use computers like a person.

The incident has sharpened internal tensions, with critics saying management dismissed exactly the safety and privacy risks that materialized. Notably, Kasriel framed the pause partly as a natural stopping point, saying Meta had now “gathered sufficient data to assess the long-term value of the tool” — leaving the program’s future, and whether the surveillance resumes, unresolved.