Meta has paused its internal AI training program, Model Capability Initiative, after sensitive employee data collected through the tool was reportedly accessible across the company, raising fresh questions around AI training, employee monitoring, privacy safeguards, and internal data security.
TL;DR
- Meta paused its Model Capability Initiative, an internal AI training program that captured employee keystrokes, clicks, mouse movements, and screen context.
- Business Insider first reported that sensitive data was accessible companywide, while Reuters confirmed Meta is investigating the exposure.
- Meta says it has no indication that employees improperly accessed the data, but the incident has intensified internal privacy concerns.
Meta has hit pause on its Model Capability Initiative, also known as MCI, an internal program built to improve AI models by collecting data from employees’ work computers, including mouse movements, clicks, keystrokes, and screen context. The move comes after Business Insider reported that sensitive employee data collected through the program was accessible across the company.
Reuters also verified the pause, reporting that Meta said it would temporarily halt the internal tracking program while investigating data security concerns. The program had rolled out in April 2026 for U.S.-based employees as part of Meta’s broader effort to train AI agents that can better understand how people complete everyday computer-based tasks.
The exposed data reportedly included private conversations, performance-related data, prompts, transcriptions, and internal sensitivity ratings. Business Insider reported that the incident was classified as a SEV 2 on Meta’s internal severity scale, where 0 is the most severe level.
Meta spokesperson Tracy Clayton told Reuters that the company had designed the program with privacy safeguards and had “no indication at this time that any data was improperly accessed.” However, Clayton said the company was pausing the program while it investigates the matter.
The controversy did not start with the leak. When MCI was first announced in April, employees raised concerns over the scope of the collection. Business Insider reported that the tool was mandatory for most staff, with Meta CTO Andrew Bosworth saying internally that there was no opt-out option for work-provided laptops.
Reuters reported at the time that MCI was intended to help Meta’s AI models learn how humans interact with computers, such as using keyboard shortcuts, selecting from dropdown menus, and completing routine workflows. An internal memo cited by Reuters said employees could help models improve “simply by doing their daily work.”
Employee pushback had already forced Meta to scale back parts of the program earlier this month. Reuters reported on June 2 that Meta planned to add controls allowing employees to pause data collection for up to 30 minutes at a time and request exemptions, after staff raised concerns about personal data, battery life, and control over when capture happens.
WIRED separately reported that potentially sensitive information from employee laptops had been left accessible internally, citing an internal security notice and current employees familiar with the issue. Engadget also noted that the pause is unlikely to calm employee concerns, given that the AI training initiative was already controversial before the data exposure.
The pause also lands during a tense period inside Meta. Business Insider recently reported that Bosworth described employee morale as among the lowest in the company’s 20-year history, following layoffs, AI-focused reassignments, and backlash to employee activity tracking.
For Meta, the issue is not just whether the MCI data was misused. It is whether the company can convince employees that its AI push, especially when built on internal work activity, can be governed with controls strong enough to match the sensitivity of the data being collected.
