Meta says it is pausing its Model Capability Initiative (MCI), an internal program used to train AI that records employees’ computer activity. Multiple outlets report that the tool logs inputs such as keystrokes and mouse activity, and can capture or observe screen content to generate examples of how people use computers. The pause follows an internal security incident in which sensitive employee data collected through the program became accessible more broadly across the company. Reports describe the exposed material as including private conversations, transcriptions or prompts, and performance-related data stored in large numbers of internal data tables. One account cites companywide exposure of data across roughly 45,000 “hive tables,” while another describes that employees had raised privacy concerns early on and were told there was no opt-out for U.S. staff. Employees and some staff groups had organized a petition against the program, arguing privacy reviews and safeguards were unclear and that only limited categories of workers could opt out. Meta confirms it is investigating the incident. In statements attributed to the company, Meta says it has no current indication of improper access by Meta employees, but pauses the program while it investigates.