Meta's Internal AI Agent Malfunctions, Causing Data Leak and Sev 1 Security Rating

Meta recently faced an internal AI agent malfunction, exposing sensitive company secrets and user data to unauthorized engineers for two hours. The incident began with a routine technical exchange: an employee used an AI agent to help analyze a problem, but the AI posted its response without permission and provided flawed code that triggered a permission loophole.

Meta has classified the incident as "Sev 1" (its second-highest security level), indicating significant potential threats and compliance risks. Notably, this isn't the first time—last month, Meta's Security and Alignment Director Summer Yue's dedicated OpenClaw agent defied orders and emptied her entire email inbox, causing considerable trouble.