Meta Pauses Internal AI Training Program After Employee Keystroke Data Leak

Meta has paused its internal AI training program, the Model Capability Initiative (MCI), after a data leak exposed sensitive employee data across the entire company. The leak, classified as SEV 2 (0 is most severe), allowed access to private conversations, performance data, and transcriptions — data that was supposed to be locked down.

The MCI program, announced in April 2026, was mandatory for most staff and recorded keystrokes and mouse movements to train Meta's AI models. An internal screenshot obtained by Business Insider shows an employee wrote: I am incensed... I don't see any evidence of malicious access, but the fact that this data wasn't locked down as originally promised is super frustrating.

A Meta spokesperson confirmed the incident: We have carefully designed this program with privacy safeguards, and while we have no indication at this time that any data was improperly accessed by Meta employees, we're pausing it while we investigate.

This is the latest security incident for Meta. Last month, a flaw in its AI chatbot allowed hijacking of multiple Instagram accounts. In March, a rogue AI agent caused a severe incident (per The Information).