Meta Pauses Unpopular 'Model Capability Initiative' Employee Surveillance Program Following Massive Data Leak
Meta has paused its controversial "Model Capability Initiative" (MCI) employee keystroke and screenshot tracking program, which was installed on U.S. employee laptops in April 2026. The pause, first enacted on June 22, 2026, followed a massive internal data breach and intense employee backlash, highlighting the severe cultural and security risks of using workplace monitoring to gather AI training data.
The Model Capability Initiative (MCI)
Under the MCI program, Meta installed monitoring software designed to automatically capture keystrokes, mouse movements, click locations, and screen content on U.S. employees' company laptops. Meta's stated purpose was to use this highly granular behavioral data to train its next-generation AI agents to operate computers.
However, the program was launched without an opt-out option for employees, and Meta failed to provide formal privacy reviews or clarify exactly how the collected data would be restricted.
The 45,000-Table Data Exposure
On June 22, 2026, Meta was forced to pause the MCI program after discovering a massive security vulnerability. As reported by Wired, employee keystroke and screenshot data collected across 45,000 internal data tables had been left accessible to the entire company. The exposed data included private employee conversations, performance records, and sensitive transcriptions.
The data leak supercharged existing worker resistance. More than 1,600 Meta employees had already signed a formal petition demanding the program's cancellation.
Leaked Audio and Executive Backlash
Following the leak, Meta CTO Andrew Bosworth addressed the pause in internal meetings. Leaked audio from a weekly meeting revealed Bosworth expressing frustration not just with the program's failure, but with employees who leaked details of the tracking software and the breach to the press, stating:
"We fine and fire leakers all the time... we're locking the whole thing down."
Bosworth defended the program's intent in subsequent public interviews, explaining that tracking employee actions was meant to teach AI how to use computers, but acknowledged that the program remains paused.
The incident highlights a growing tension between big tech's insatiable demand for high-quality human interaction data to train AI agents and the fundamental privacy rights and trust of their own workforces.