Meta has paused its initiative to monitor employee keystrokes and mouse movements for AI training, just two months after announcing the program. The decision follows an alleged internal leak that exposed sensitive employee data across the company.
According to a screenshot obtained by Business Insider, the exposed data included employees’ private conversations, performance data, and transcriptions. The incident was classified as a SEV (severity) 2 on a scale where 0 is most severe. Documents reviewed by Wired revealed a security notice stating that “employee data across 45,000 hive tables” were exposed, encompassing “full prompts and transcriptions, private conversations, people and performance data.”
The program, internally called the Model Capability Initiative (MCI), was designed to train AI models. A Meta spokesperson told Fast Company: “We have carefully designed this program with privacy safeguards and while we have no indication at this time that any data was improperly accessed by Meta employees, we’re pausing it while we investigate.”
Meta employees had strong reactions to the MCI program, with staffers reportedly unable to opt out. The pause reflects growing internal and external scrutiny of workplace surveillance, especially when used to train AI systems. This incident underscores the tension between ambitious AI training data collection and employee privacy.
The leak raises broader questions about corporate governance of AI data sourcing. As companies race to gather training data, internal safeguards may not keep pace with employee expectations of privacy. Observers will watch whether Meta reinstates the program or shifts its approach entirely.