Overview
- Meta paused the Model Capability Initiative after engineers filed a high‑priority SEV and flagged that MCI data had been exposed companywide on Monday.
- The MCI tool, rolled out in April, recorded U.S. employees’ mouse movements, clicks, keystrokes and occasional screen snapshots to create training data for AI models.
- Internal notices and documents say the exposure touched roughly 45,000 hive tables and included full prompts, transcriptions, private conversations and people and performance data.
- Meta says investigators have found no confirmed improper access so far and that the program will only be reenabled once data protections meet the company’s standards, and the CTO acknowledged the rollout fell short of its privacy review.
- Employees had already protested the mandatory program, with more than 1,600 signing a petition, and the incident deepens broader morale and governance concerns as Meta pushes to accelerate its AI work.