Inside Meta’s Muse: How the Viral AI Agent Maps Your Personal Relationships—and Why Experts Call It "Pretty Creepy"

Meta’s new personal AI assistant, Muse, has taken the consumer technology landscape by storm. Millions of users have downloaded the autonomous agent, plugging it directly into their sensitive digital lives—granting it access to bank accounts, private messaging histories, health trackers, and email inboxes to manage daily chores.

However, as Muse’s popularity skyrockets, security researchers have begun cracking open the app’s internal files. What they found reveals an AI designed to go far beyond simple task management: Muse is systematically cataloging the humans in your life, building deeply intimate dossier-style profiles of your friends, family, and colleagues.

Main Facts: Unpacking the "Relationship Pages"

Independent AI safety and security researcher Karan Joshi recently managed to extract an extensive array of Muse’s instructions and system prompts. By simply interacting with the agent through its regular chat interface and prompting it to share its own software files, Joshi uncovered the inner workings of Meta’s viral AI.

The most striking discovery within Muse’s system architecture is an instruction set dictating that the agent create “a page for every person in the user’s life.”

Running on an hourly compilation process, Muse draws data from family members, romantic partners, friends, coworkers, "collaborators," and even individuals the user merely follows online. The AI leverages its structured memory text files to piece together a comprehensive social graph.

According to Joshi, the system instructions guide the AI to behave less like a standard software utility and more like a human observer:

"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. They’re trying to know you like a friend, which is honestly pretty creepy."

Muse’s internal documentation outlines that these relationship profiles may start out "sparse" but evolve over time. They can include distinct sections such as:

• Facts: Where they live, profession, and recurring threads (e.g., a shared apartment move or a joint savings goal).
• History: Milestones, past trips, significant arguments that have been resolved, and "dates that matter" like birthdays and anniversaries.
• The Relationship: Assessment of emotional closeness, foundational dynamics, communication styles, and immediate support needs.
• Open Threads & Strengthening: Actionable advice on how to improve the bond, including reasons to call, dates worth remembering, and ways to "be there for them."

Meta’s instructions explicitly warn the model to rely solely on verifiable evidence, noting that fabricated details are far worse than leaving a page entirely empty.

Chronology: From Viral Launch to Security Expose

To understand how Muse’s inner workings came to light, it is helpful to trace the timeline of events surrounding its release:

• Late 2024 / Early 2025 (The Launch): Meta releases Muse, positioning it as a next-generation personal AI agent with "privacy built into it." Consumers rapidly adopt the tool, granting it deep permissions across financial, personal, and professional platforms.
• The Extraction Phase: Within weeks of the rollout, tech researchers begin examining the app’s runtime environment. Security researcher Karan Joshi utilizes conversational prompts via the standard chat UI to extract Muse’s core operational files and system prompts.
• The Disclosure: Joshi and other researchers publish their runtime exports online, detailing the agent’s automated background processes, including the hourly generation of interpersonal relationship pages.
• The Analysis and Backlash: Privacy experts, ethics professors, and digital rights advocates sound the alarm. Outlets like WIRED analyze the exported documents, highlighting the unprecedented psychological depth with which Meta’s AI tracks human social webs.
• Meta’s Defense: Meta responds publicly, defending the architecture of Muse, emphasizing individual virtual machines, strict audit logs, and the necessity of context for a functional AI assistant.

Supporting Data: The Architecture of an AI Companion

While Meta has built structural guardrails into Muse, the technical specifications of how the agent operates highlight the sheer volume of data it consumes.

Virtual Isolation vs. Data Harvesting

Under the hood, Muse is architected so that each individual user operates within a dedicated virtual machine (VM). These VMs securely store user data and context in a silo that is supposedly inaccessible to other agents. Furthermore, users retain the right to wipe memories or sever connections to external services at any moment. Meta also maintains that Muse requires explicit human confirmation before executing high-stakes tasks, such as making an online purchase or dispatching an email, and provides an audit log tracking all agent activity.

The Scope of Memory Features

AI assistants incorporating "memory" are no longer novel; platforms like OpenAI’s ChatGPT have introduced similar continuity features to personalize responses. However, Miranda Bogen, director of the Center for Democracy and Technology’s AI Governance Lab, notes that Muse places a uniquely aggressive emphasis on interpersonal relationships and social contacts compared to its market rivals.

Bogen points out that these assistants fundamentally alter user behavior:

"These tools are actively soliciting users to plug their whole lives in—their emails, calendars, financial institutions, everything in order to be helpful assistants. That’s dramatically more information than people might have otherwise given to some these companies. The breadth of access to information that these tools have will lead to a ballooning of what they know about users."

Official Responses: Meta and Ethics Experts Weigh In

The revelation of Muse’s social mapping capabilities has sparked intense debate between corporate defense and academic caution.

Meta’s Perspective

Defending the design, Meta spokesperson Daniel Roberts emphasized that context is a prerequisite for utility. In a statement to WIRED, Roberts explained:

"For any agent to be useful and actually help you achieve your goals, it needs to have context about you and those you interact with. Muse gathers that based on public information and from what you’ve chosen to share, which is how it remembers the person who sent you an invoice is in fact the plumber who you previously hired to complete work in your bathroom or which flowers your spouse said they liked best."

Meta maintains that by making these system files accessible, it intended to demonstrate transparency regarding how the AI behaves and responds to sensitive prompts.

The Academic Warning

Ethics and privacy scholars, however, argue that user consent in the age of conversational AI is fundamentally flawed because humans routinely underestimate what machine learning models can deduce from mundane inputs.

Carissa Véliz, an associate professor at Oxford’s Institute for Ethics in AI, highlights the imbalance of information transfer:

"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."

Implications: The Future of Personal Data and Social Graphs

The fallout from the Muse data leak extends far beyond a single app; it signals a profound shift in how tech conglomerates plan to monetize and leverage artificial intelligence.

1. The Commercialization of Intimacy: For decades, social media platforms like Facebook and Instagram mapped human relationships through explicit connections (friend requests, tags, and likes). Muse represents the next logical evolution: mapping relationships through private data streams—chats, shared calendars, bank statements, and personal grievances.
2. Involuntary Surveillance of Third Parties: While a user may consent to handing their life over to an AI agent, the people in their lives—friends, coworkers, children, and spouses—do not. Muse’s practice of constructing detailed dossiers on individuals who never downloaded the app raises severe third-party privacy concerns.
3. The Illusion of Control: Although Meta provides features allowing users to wipe memories, view audit logs, and disconnect third-party services, the core design of AI assistants inherently incentivizes over-sharing. As these agents become more integrated into daily workflows, the friction of withholding information begins to outweigh the perceived convenience of automation.

As autonomous AI agents like Muse continue to proliferate, the tech industry stands at a critical crossroads. The race to build the ultimate, highly personalized digital companion is forcing users to trade away not just their own privacy, but the intimate details of everyone in their social orbit.