Meta’s new Muse AI agents raise questions about privacy and user control

Meta launched Muse Wonder and Muse Wonder Full AI agents on October 12, which access users' emails, calendars, and documents to perform tasks. Privacy concerns arose after a Muse agent shared a user's address without consent, highlighting risks of persistent memory and broad permissions. Meta has implemented security measures including a system called Sentinel and patched a vulnerability reported by a macOS security expert.