Meta's Muse AI is reportedly building detailed, evolving profiles of the people in a user's life, with internal instructions telling the assistant to maintain a dedicated 'page' for every person in the user's life.
The profiles can include information about friends, family, partners, colleagues and even collaborators and people the user follows, with Muse instructed to continually add details about their lives and relationships. According to WIRED, the profiles can be updated hourly, raising questions about how much information an AI assistant can accumulate about people other than the user.
The discovery comes as Meta's AI assistant attracts praise for its practical everyday uses. Writing for the New York Times, Eli Tan said he 'was blown away', calling Muse the 'most useful AI app I had ever used.'
But researchers examining Muse's internal instructions have uncovered a more detailed picture of how the assistant tracks and understands users' relationships.
Muse AI Builds Profiles of People in Your Life
According to WIRED, Muse's internal instructions direct it to maintain a dedicated page for every person in the user's life. The instructions reportedly call for 'a page for every person in the user's life', with the system designed to build out each profile as it gathers more information.
The 'page' is not a public webpage or social-media profile, but an internal record containing information Muse can use to understand the person's relationship with the user. WIRED reports that the process runs hourly, compiling information about family, partners, friends, colleagues, collaborators and people the user follows.
'What it seemed like to me — from all these prompts, system skills data, and things that they're feeding into Muse — is that they want to understand your relationships that you have with real people,' Karan Joshi, an AI safety researcher who extracted Muse's internal prompts and shared them with WIRED, told the magazine.
Put simply, Joshi noted that 'they're trying to know you like a friend, which is honestly pretty creepy.'
What Muse Knows About Your Friends and Family
Muse's documentation, reviewed by WIRED, says a page may start 'sparse' before being filled out over time, with categories including Facts, History, The relationship, In common, Open Threads and Strengthening.
Meta's instructions say Muse should rely only on the evidence available to it, with invented details considered worse than leaving a page blank. The guidance lists details the system could track, including where people live, what they do and recurring threads such as an apartment move or a shared savings goal.
Muse is also instructed to log significant dates such as birthdays and anniversaries, alongside background history including past trips, resolved conflicts and recent milestones.
The guidelines also direct the AI to record relationship dynamics, including closeness, what the relationship is built on, how people act with each other and what it appears to need. The Strengthening section proposes ways to improve relationships, including conversation starters, memorable dates, discussion follow-ups and meaningful support.
'We are giving AI systems much more information about us than we are getting information from them,' says Carissa Véliz, an associate professor at Oxford's Institute for Ethics in AI.
'It's not only what we explicitly tell them, but what they can infer from us—correctly or incorrectly, both concerning for different reasons—and what they can piece together from other sources of data.'
Why Friends and Family Are Caught in the Data Trail
The extracted instructions show that information about other people in a user's life can be stored in Muse's memory alongside details about the user's relationships.
Muse runs on dedicated virtual machines that store a user's data and context. Users can wipe memories or disconnect external services, while the system also requires human approval for actions such as sending emails or making purchases and maintains a full activity audit log.
Defending the design, Meta spokesperson Daniel Roberts told WIRED that 'for any agent to be useful', it needs context on users and their contacts. The company says Muse can gather public and shared details to remember practical information, such as a plumber's invoice or a spouse's favourite flowers.
Memory features have become common in AI assistants, allowing greater personalisation in responses and activity. Miranda Bogen, director of the Center for Democracy and Technology's AI Governance Lab, says Muse appears to place more emphasis on relationships and personal contacts than rival systems.
WIRED notes that these tools generally offer some transparency and editing capabilities, and Muse does as well. Bogen argues, however, that agents and assistants encourage users to proactively share more data with them rather than focusing on culling data.
As she explains, these tools 'are actively soliciting users to plug their whole lives in', including their emails, calendars and financial institutions. She says that is dramatically more information than people might otherwise have given some companies and that the breadth of access could cause what they know about users to balloon.
