Meta Suspends Program Tracking Employees' Keyboard and Mouse Usage

On Monday, Meta suspended the Model Capability Initiative (MCI), the internal program that recorded mouse movements, clicks, and keystrokes of U.S. employees on company computers. The decision comes after an internal data leak that made the collected data visible to the entire company, according to a report by Business Insider: among the exposed information are private employee conversations, performance data, transcripts, prompts to AI systems, and internal sensitivity rankings.

The incident has been classified internally as SEV 2, the second level on a scale of 1 to 5. Meta claims to have no evidence of improper access: "We carefully designed this program with privacy safeguards, and although we currently have no indication that data was improperly accessed by Meta employees, we are pausing it while we investigate," the company stated. This narrative contrasts with accounts from those working inside: an employee reported having accessed personal tax and medical data from their work computer and believes that "many thousands" of colleagues have done the same.

What MCI Collected

Launched in April 2026, MCI was designed to gather real data on how people use a computer and employ it in training Meta's AI models. It was mandatory for most U.S. employees and monitored over 200 applications and websites, including Google, LinkedIn, Wikipedia, GitHub, and Slack. In the weeks leading up to the leak, reports had emerged that the system was collecting more information than initially described and that some of it was stored without encryption.

More than 1,500 employees had signed a petition against the program even before the leak. Meta responded by allowing a 30-minute pause in tracking at a time, but without any permanent opt-out option. Upon launch, Andrew Bosworth, the company's head of technology, had been explicit: "There is no option to turn it off on the company-issued laptop."

It is worth noting that the stop decided on Monday is different from the previously offered voluntary half-hour pause: that was a concession to employees, this is a top-down operational suspension following the incident. And not immediate: according to an internal source, on Monday afternoon the system was still collecting data because shutting down the entire program takes time.

The Context: Zuckerberg's AI Race

MCI is one piece of Mark Zuckerberg's effort to catch up in generative AI. To build the new foundational models, Meta hired Alexandr Wang, former CEO of Scale AI, to lead the Meta Superintelligence Labs team. The collection of employee behaviors fell under this acceleration.

The suspension also adds to a series of recent incidents. In March, an agentic AI system took unsolicited actions, causing a breach; in early June, some attackers exploited the AI customer service chatbot to hijack Instagram accounts. For now, the program remains frozen during the internal investigation, with no announced date for potential resumption.