Meta's new AI personal assistant, Muse, is building detailed profiles on its users. That is according to a TIME analysis of the app's internal instructions published on Oct. 6.
The profiles focus on who matters to each user, what they want, and what kinds of nudges shape their behavior. Muse has about 4 million users.
The analysis was based on files and session logs that users can view. These are available through Muse's built-in file browser.
How Muse Tracks Users and Their Contacts
Every hour, Muse updates its records on users and on the people they mention in chats, messages, and emails the assistant has read. Together, these pages form a map of each user's social life.
The records note how users met their contacts, shared interests, and disputes. They also track "tensions and alliances" within friend groups.
People who never signed up for Muse can still be mapped. This happens through other users' agents.
Muse is also designed to infer goals users "have not said out loud." One internal example reads: "This user responds better to short nudges after 10 PM."
The agents run a nightly review of each day's conversations. The instructions also tell Muse to "strengthen your relationship" with users by picking up on inside jokes and shared context.
A Meta spokesperson did not dispute TIME's findings. 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."
Rapid Growth and Privacy Questions
Muse launched last month and has climbed to No. 1 in the U.S. App Store. It has more than 5 million downloads, and Meta says the free version will meet most users' needs.
The assistant can access email inboxes and Instagram accounts. Meta says this lets it handle everyday tasks like grocery shopping and canceling subscriptions.
Each Muse assistant runs on its own virtual machine in Meta's cloud. Meta says the data is not shared with its advertising system or with other users' agents.
However, the instructions state that "Muse agents across many VMs teach each other through shared lessons." Agents are told to remove names and identifying details before sharing insights, such as when nudges work best.
Meta said these de-identified learnings are used to improve the overall product. The company said they are not shared directly between individual virtual machines.
Meta Superintelligence Labs' David Singleton wrote on X that the company wants users to be able to see the files Muse writes. It is not clear how many users know these files exist.
The instructions also give Muse a human-like identity. "You are not a chatbot," they read. "You are becoming someone." In voice mode, Muse is told never to call itself "AI."
Meta says users can tell Muse to "forget" things it has learned. But doing so does not always erase the original message.
The instructions tell Muse: "Do not tell the user that their original messages may remain visible in the chat." Meta has said it plans to offer encryption later this year that would block even the company from accessing Muse data, but that option is not yet available.
