Meta pauses internal AI training program after employee data leak

Meta halted an internal AI training initiative that collected keyboard and mouse data from its staff after a leak exposed sensitive employee information company‑wide. The breach revealed private conversations, performance metrics and transcripts, and was classified as a SEV‑2 incident. Meta said there is no evidence of malicious employee access and the program – known as the Model Capability Initiative – has been suspended while security and privacy controls are reviewed.

Employees voiced frustration, with one remarking, "I am very angry," over what they said were insufficient safeguards. The pause underscores growing industry concerns about using real‑user interaction data for AI model development and the need for tighter data‑access governance.