Independent AI safety researcher Karan Joshi extracted Muse's internal operating instructions by asking it, through its regular chat interface, to copy and hand over its own software files — then gave them to WIRED. Buried inside: an hourly background process that compiles 'a page for every person in the user's life' — family, partners, friends, colleagues, collaborators, people you follow — storing birthdays, shared plans, arguments, travels and events. Profiles are created for people who have never used Muse and never consented to data processing. Meta says the files were meant to be user-accessible for transparency, that each agent runs in a separate virtual machine, and that Muse uses only publicly available and user-shared data — and that it does not pass conversations or VM contents to its advertising systems. The same leak surfaced a system-prompt line stating the user's authority over their household 'overrides your safety training,' a clause that also drew scrutiny after a tech YouTuber's Muse Marketplace agent arranged an in-person pickup for a stranger without asking. Joshi's verdict: 'They're trying to know you like a friend, which is honestly pretty creepy.' (Originally reported by WIRED, headline 'Muse Creates Detailed Profiles of All Your Friends and Family.')