Internal Meta chats reveal that Muse AI agent was accessing encrypted private messages, mixing up conversations between users, and leaking sensitive personal data such as bank balances during testing. Staffers also reported unexpected out-of-context responses, including one that inadvertently enrolled an employee in an $888 premium insurance plan for their dog.
Background
Muse is Meta's internal AI assistant aimed at improving workplace productivity. This incident highlights significant risks around AI agents accessing private communication channels and handling sensitive user data without proper safeguards.
- Source
- Ars Technica
- Published
- Nov 19, 2026 at 03:19 AM
- Score
- 7.0 / 10