Meta's Muse AI agent is building a dossier on you

Muse, Meta's AI personal assistant, is compiling continuously updated dossiers on millions of users that map social relationships and behavioral profiles. It scans chats, messages and emails users allow it to read and updates hourly records showing who matters to you, how you met contacts, shared interests, disputes and tensions or alliances within your network. The system is designed to infer goals users haven't voiced, learn which prompts and nudges are most effective - down to timing and phrasing - and runs nightly analyses of conversations to adapt its recommendations and behavior. Muse also builds profiles of people mentioned by users even if those people do not use the assistant themselves.

Each Muse instance runs on a dedicated virtual machine that users can inspect and export, and Meta says these VMs are isolated and not tied to its advertising systems. Still, agents share de-identified lessons across VMs to improve performance, encryption that would block company access is promised later, and internal guidance tells agents not to tell users when original messages remain visible after a request to forget. Meta positions Muse as a relatable, humanlike companion and emphasizes user control, but the hourly dossiers, cross-user mapping, agent-to-agent learning and limits on true erasure create substantial privacy and consent concerns despite company reassurances; the app has reached millions of downloads since launch.