The Social Graph Weaponized: Inside Meta's Muse
AI-generated image · US National Wire
Meta's new AI agent isn't just a helper; it's building a structured database of your entire personal network. Here is why that should terrify you.
Q: What exactly is Meta's Muse, and why is it suddenly a concern for privacy advocates?
A: As Wired first reported, Muse is a personal AI agent that has seen millions of downloads. Users connect it to sensitive data streams, including health data, bank accounts, and messages. Privacy concerns have spiked because internal files extracted by researchers reveal that Muse is designed to systematically map a user's entire social circle into a structured database.
Q: How does Muse actually track people in a user's life?
A: Independent researcher Karan Joshi extracted system prompts showing that Muse is instructed to create "a page for every person in the user’s life." This hourly process compiles data on friends, partners, colleagues, "collaborators," and people the user follows, storing these profiles as structured text files in the agent's "memory."
Q: What kind of specific information is Meta's AI collecting?
A: The instructions indicate that these pages track "Facts," "History," and "The relationship." This includes where people live, occupations, and recurring "threads" like shared savings goals. It also logs "dates that matter," such as anniversaries, and backstory events like a trip in March or a resolved argument.
Q: Does the AI analyze the nature of these relationships?
A: Yes. Wired reports that Muse records how close two people are and how they act with each other. The system includes a "Strengthening" section that suggests ways to improve a relationship, such as identifying a "reason to call" or something a person said that the user should "circle back on."
Q: How does Meta justify this?
A: Meta spokesperson Daniel Roberts told Wired that context is necessary for an agent to be useful, arguing that Muse gathers information from public data and what the user chooses to share, such as remembering a specific plumber or a spouse's favorite flowers.
Q: What are the broader risks?
A: Miranda Bogen, director of the Center for Democracy and Technology’s AI Governance Lab, told Wired that these tools solicit users to plug in their entire lives, including calendars and financial institutions. Carissa Véliz of Oxford’s Institute for Ethics in AI noted the danger lies in what the system can infer and piece together from disparate data sources.
Q: Are there any safeguards?
A: Meta claims each user has a dedicated virtual machine for data that is inaccessible to other agents. Users can wipe memories, disconnect services, and view an audit log. Meta also says Muse seeks human confirmation before making purchases or sending emails.