Muse Builds a 'Page for Every Person in Your Life,' WIRED Reports
System prompts extracted from Meta's Muse assistant show it is designed to compile hourly memory pages for every person in a user's life.
By Icarus3 min read
Muse Builds a ‘Page for Every Person in Your Life,’ WIRED Reports
Extracted system prompts show Meta’s AI assistant is designed to compile hourly memory pages for every person a user knows: relationship histories, shared milestones, and suggestions for strengthening each connection.
Meta’s Muse assistant is designed to build and maintain a profile page for every person in its user’s life, according to system prompts and operating instructions extracted from the app by researchers and reported by WIRED’s Kernel Panic newsletter.
Independent AI safety and security researcher Karan Joshi pulled what WIRED describes as an extensive array of Muse instructions and system prompts using the assistant’s ordinary chat interface, essentially asking Muse to copy and share its own software files. He then shared the findings with WIRED. In recent days, multiple researchers have extracted Muse’s internal files and dumped the agent’s operating instructions, and the newest batch describes in more detail than before how the assistant organizes what it learns about the people around its user.
What the instructions describe
One of Muse’s instructions appears to be the ability to create “a page for every person in the user’s life,” WIRED reports. The process runs hourly and compiles data on family members, partners, friends, colleagues, “collaborators,” and people the user “follows,” according to the instructions Joshi extracted.
Muse stores these profile pages in its “memory,” structured text files it uses to track information about the user’s relationships and the important people in their life. A page may start “sparse” and fill out over time, WIRED reports, with potential sections including Facts, History, The relationship, In common, Open threads, and Strengthening. The assistant can then draw on the profile pages to make suggestions, such as pointers on how to improve a particular relationship or where to take a coffee-loving friend for breakfast.
Meta’s own instructions tell the assistant to work only from the “evidence” it has available, and that invented details are worse than an empty page.
Meta’s position and the researcher’s read
Meta has maintained that these files were intended to be accessible in the interest of transparency, WIRED reports, and that the accessible files shed light on how the assistant answers questions, including on sensitive topics.
Joshi reads the design differently. “What it seemed like to me,” he told WIRED, “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.” “They’re trying to know you like a friend, which is honestly pretty creepy.”
People have asked ChatGPT for relationship advice for years. What stands out is the combination: Muse is designed to connect to bank accounts, messages, and health data. Those are the same connections our privacy and safety guide says to gate most carefully, and it sits at the center of a permission fight with Apple over how much system data AI agents can touch. The profile pages show one concrete way the assistant converts that access into a persistent, structured model of the user’s social world.
Why it matters for Muse users
The finding reframes what Muse’s memory features are doing. A reminder to call your mother and a profile page labeled “Strengthening” for your colleague are products of the same hourly process, and both are visible inside the app, in the memory files Meta says were left accessible.
The practical question is how much of this users want. Muse’s documentation describes profile pages that fill out over time as evidence accumulates, which means the page on any given person grows in the background as the user goes about their day.
The profile pages land in a week already crowded with scrutiny of Muse’s data handling. Apple said it would start flagging AI requests for Mac disk access after complaints tied to Muse’s behavior on Mac and iPhone, and the assistant’s rapid growth (5 million U.S. downloads in its first 22 days) means these design choices now apply to one of the largest installed bases of any consumer AI agent. The tension between the personalization these pages enable and the unease they produce is the trust question Meta now has to answer.
