Meta Muse AI Creates Detailed Profiles of Your Friends and Family
Meta’s new personal assistant, Muse, has quickly become a viral hit with millions of consumer downloads, but newly extracted internal files reveal that the AI agent runs an hourly process to create detailed profiles of every person in a user’s life. According to reporting from WIRED, independent AI safety and security researcher Karan Joshi extracted the operating instructions and system prompts by prompting the chat interface to share its own software files, revealing that Muse compiles data on family, partners, friends, colleagues, and collaborators into structured text files.
The Tech TL;DR: What Muse’s System Prompts Reveal
- Hourly Profiling: Muse runs background routines to build and update dedicated pages for friends, family, and colleagues based on user interactions.
- Extracted Operating Instructions: Independent security researcher Karan Joshi extracted the system files directly through the chat interface and shared the findings with WIRED.
- Granular Data Points: The structured text files track specific personal details, including shared history, recurring threads, important dates, and relationship-strengthening suggestions.
How Muse Compiles Personal Relationship Profiles
As millions of users connect the AI agent to bank accounts, messages, and health data to complete everyday tasks, the internal files exposed by Joshi show how the system structures social information. WIRED reported that Muse’s documentation outlines an hourly process where the assistant builds “a page for every person in the user’s life.” These files start sparse and expand over time into specific categories: Facts, History, The relationship, In common, Open threads, and Strengthening. Meta’s instructions direct the model to rely strictly on available evidence rather than inventing details, logging information such as recurring apartment moves, shared savings goals, birthdays, anniversaries, and past arguments or milestones.
Contrasting Perspectives on Surveillance and AI Memory
While AI chatbots have long handled basic relationship inquiries, security and privacy researchers point out that Muse’s architecture represents a significant escalation in persistent profiling. “What it seemed like to me—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,” Joshi told WIRED, describing the capability as “honestly pretty creepy.” Dnyuz corroborated these findings, noting that the system instructions prompt the model to assess relationship dynamics, such as “how close they are, what it is built on, how they act with each other, and what it seems to need right now.” Carissa Véliz, an associate professor at Oxford’s Institute for Ethics in AI, told Dnyuz, “We are giving AI systems much more information about us than we are getting information from them. It’s not only what we explicitly tell them, but what they can infer from us—correctly or incorrectly, both concerning for different reasons—and what they can piece together from other sources of data.” Meta has maintained that the internal files were intended to be accessible for transparency, offering a clear window into how the agent processes sensitive and interpersonal queries.
Isolated Virtual Machines Expose Foundational Runtime Instructions
Under the hood, Muse is architected so that each individual user operates within a dedicated virtual machine designed to store personal data and contextual logs. Dnyuz reported that this isolated virtual machine architecture handles the heavy lifting of maintaining structured text file memories across sessions. Joshi’s discovery highlighted that users interacting with the default chat interface could extract these foundational runtime instructions simply by asking the agent to share its own files, demonstrating a unique transparency—or vulnerability—in how consumer-facing AI agents expose their core system prompts to end users.