Meta's Muse Is Building Dossiers on Your Friends and Family, Even If They Never Signed Up, Internal Files Reveal

Meta's Muse may be marketed as a helpful personal assistant, but internal instructions reviewed by TIME and extracted by an independent researcher indicate that the AI agent is designed to build detailed, continuously updated profiles of users and people in their lives, including people who have never used Muse themselves.

Muse has grown rapidly since its 8 September launch, with more than 4 million people using the assistant in some way at least once a week, according to internal figures reviewed by The Information. The findings raise questions about how much an AI assistant can learn about people who never chose to use it.

Muse Is Mapping Your Social Circle

TIME's analysis found that Muse's internal instructions call for updating these profiles roughly every hour, using information about people mentioned in chats, messages, and emails the assistant can access. The profiles can include how users met their contacts, shared interests, disagreements, and 'tensions and alliances' within their social circles.

WIRED separately reported that Muse's instructions call for creating 'a page for every person in the user's life,' including family, partners, friends, colleagues, collaborators, and people the user follows. As a result, people who have never used Muse could still appear in a user's relationship profiles if information about them is available to the assistant.

AI safety researcher Karan Joshi, who extracted Muse's internal instructions and shared them with WIRED, said the system appeared designed to understand users' real-world relationships. 'They're trying to know you like a friend, which is honestly pretty creepy,' Joshi told the publication.

It Goes Beyond Names and Relationships

Muse is also instructed to infer users' goals, including ones they 'have not said out loud,' according to the internal material. Its internal instructions reportedly show that it can identify which prompts work best for individual users. One example says, 'This user responds better to short nudges after 10 PM.' The instructions also direct Muse agents to perform a nightly analysis of the day's conversations with the user.

Asked about the process, a Meta spokesperson did not dispute the findings, instead citing the company's blog. The spokesperson said Muse 'remembers what matters most to you, including information about others that you choose to share, so it can be a helpful personal assistant.'

Meta Says Muse User Data Is Kept Separate

Meta says each Muse runs on a dedicated virtual machine, with users controlling which information and services the assistant can access. According to the company, Muse does not share conversations or data stored in a user's VM with its advertising systems, while individual VMs are isolated from one another.

However, TIME found instructions stating that 'Muse agents across many VMs (virtual machines) teach each other through shared lessons.' Meta told the outlet that these learnings are de-identified before being used to improve Muse and that information is not directly shared between individual VMs.

Meta plans to introduce Muse Confidential VM later this year, with the entire VM, including user data and conversations, encrypted using a key only the user holds. The company said in its September announcement that the feature would be introduced later in 2026, while it is reported on 7 October that it was not yet available.