Meta Muse Contact Pages: What WIRED Found

Correction, 5 October 2026: An earlier version stated that Muse automatically maintains detailed dossiers on every person in every user’s life and described an Apple security-policy change as a response. Those assertions went beyond the available evidence. This article now distinguishes an extracted design instruction from observed behavior and removes the unsupported Apple claim.

Internal instructions extracted from Meta’s Muse assistant describe a background process that can maintain a page about people connected to a user, according to a 3 October WIRED investigation. The documents describe how the agent might organize details about family, friends and colleagues. They do not prove that every user has such pages or that Muse has accessed information a user did not share or connect.

What the extracted instructions say

Researcher Karan Joshi asked Muse through its normal chat interface to share internal files, then gave those files to WIRED. One instruction refers to an hourly process and a page for each person in the user’s life. The described template includes Facts, History, The relationship, In common, Open threads and Strengthening. It tells the assistant to keep a page sparse when evidence is thin and not invent dates or details.

The finding is about the system’s instructions, not an independent census of real profiles. WIRED reported examples of the kind of information the template could hold: where someone lives, shared events, birthdays and suggestions for staying in touch. Another review of exported Muse files describes similar relationship-memory instructions, but the evidence is not a measurement of how often the feature runs for users.

Why third-party data matters

Meta says Muse can connect to email, calendars and other apps and run tasks on a dedicated virtual machine. A person who sends an email to a Muse user might have information about them present in a connected inbox. That creates a privacy question for contacts who never signed up for Muse. It does not, by itself, establish that Muse secretly reads every private message or creates a complete record about every contact.

Meta spokesperson Daniel Roberts told WIRED that Muse uses public information and information the user chooses to share. Meta says users can wipe memories, disconnect services and review an activity log. Those are company claims about available controls; WIRED’s report did not independently verify every deletion or retention scenario. Meta also describes a separate safety process for approving sensitive actions in its security explanation.

What users can check

Users who have connected Muse to personal accounts can review the connected services and the assistant’s memory and activity controls. They can remove a connector they no longer want the agent to use. The practical issue is the breadth of information an agent can encounter after a user grants access; the exact information available depends on the services and permissions that user chose.

Meta introduced Muse in September as a personal AI agent. Our earlier coverage of Muse for small businesses concerns a separate product announcement. The relationship-page finding is a distinct privacy investigation, first reported by WIRED on 3 October.