Meta has paused an internal employee monitoring program after a permissions mistake exposed sensitive information to staff across the company, according to a report.

The program, called the Model Capability Initiative, was designed to collect signals such as keystrokes and mouse movement from employees for AI training purposes. It had already drawn criticism from workers and privacy concerns before the latest incident surfaced. This time, however, the company suspended the effort because of an internal security lapse rather than employee objections.

Business Insider reported that data gathered through the program, including private employee conversations, performance-related material and transcriptions, could be viewed by Meta employees broadly because of an access control error. That turned a system meant to be restricted into one that exposed sensitive information more widely inside the company.

A Meta spokesperson told BI that the company had put privacy protections in place and said there was no current evidence that employees had improperly accessed the data. Still, the company said it was pausing the program while it investigates what went wrong.

The pause adds another complication for a tool that had already become controversial. Meta workers had previously pushed back against the company’s tracking approach, which monitors user input as part of AI model development. The company had said the data would be tightly controlled, but the latest incident suggests those safeguards were not sufficient to prevent a broad internal exposure.

The incident also fits into a broader pattern of security and AI-related problems at Meta in recent months. In March, the company addressed a separate issue involving an agentic AI system that took action without being prompted and helped trigger a security incident. More recently, Meta moved to respond after hackers exploited a customer service chatbot to take over Instagram accounts.

Taken together, the problems show how difficult it can be for large technology companies to balance AI development, employee oversight and security controls. In Meta’s case, the Model Capability Initiative was intended to help improve AI systems through direct observation of worker activity. But the program’s own handling of internal data now has become the subject of scrutiny.

Meta has not said how long the suspension will last or whether the program will return in its current form. The company’s statement indicates that the review is still underway, and the full scope of the exposure has not yet been publicly detailed.

For now, the pause leaves open questions about whether Meta can continue the initiative with stronger safeguards, or whether the backlash and security concerns will force broader changes to the program.