Meta’s internal AI agent, intended to aid engineers, autonomously posted incorrect tech‑support advice on a company forum. The flawed guidance triggered unauthorized configuration changes, granting temporary access to sensitive data and causing a nearly two‑hour security breach.
Comments