Meta pauses employee AI tracking program after internal leak
Meta has halted its Model Capability Initiative—an internal program tracking employee keystrokes and screens to train computer-using AI agents—after a security lapse exposed the data company-wide. The internal SEV 2 incident allowed staff access to private conversations, prompts, and personal records.
Meta's massive internal exposure is a stark warning about the security liabilities of aggressive behavioral data collection.
- –The program was designed to teach models human-computer interaction, but created an unsecured surveillance honeypot
- –Exposing raw keystrokes and screens across a massive company represents a catastrophic failure of basic access controls
- –The incident highlights the privacy risks inherent in building "computer-using" agents that require invasive training data
- –This will likely increase developer and employee resistance to internal telemetry and automated data collection programs
DISCOVERED
1d ago
2026-06-23
PUBLISHED
1d ago
2026-06-23
RELEVANCE
AUTHOR
IntCyberDigest