MikhbarMIKHBAR
Artificial Intelligence

Meta's Muse AI Agent Creates Detailed Profiles on Contacts

Meta’s personal AI assistant has quickly drawn millions of users, but internal files reveal an automated process that builds structured background profiles on the people in a user's life.

Meta's Muse AI Agent Creates Detailed Profiles on Contacts

The Rise of Meta's Personal Assistant

Meta's new personal assistant, known as Muse, has quickly become a viral hit among consumers worldwide. Millions of users have downloaded the AI agent, connecting it directly to their bank accounts, messaging apps, and health data to allow the system to complete everyday tasks.

As the software takes off among consumers, detailed data from inside the mobile application is providing fresh insight into how the system organizes, processes, and presents information back to its users.

Extracting System Prompts and Instructions

In recent days, multiple independent researchers have managed to extract Muse’s internal files. By using the regular chat interface to ask the system to copy and share its own software files, security researchers gained a rare glimpse into how the underlying architecture behaves.

Independent AI safety and security researcher Karan Joshi successfully extracted an extensive array of system prompts and instructions. Joshi subsequently shared these comprehensive findings with WIRED, shedding light on the automated routines operating beneath the surface.

Automated Relationship Profiles

Among the instructions discovered in the extracted files is a capability for the AI agent to create a dedicated page for every person in a user's life. This hourly background process involves compiling information on family members, romantic partners, friends, colleagues, collaborators, and individuals the user follows.

The overarching goal is for the assistant to leverage its internal memory architecture to track social information and relationship dynamics. The system can then generate tailored suggestions, such as offering pointers on improving specific relationships or recommending a breakfast spot for a friend who loves coffee.

Inside the Structured Data Pages

Documentation indicates that these individual contact pages start out sparse and are gradually filled out over time. Potential sections mapped out in the instructions include basic facts, shared history, relationship context, commonalities, open threads, and strengthening suggestions.

Meta's instructions explicitly state that the model should rely only on available evidence, noting that invented details are worse than an empty page. The software can record locations, professions, recurring conversation threads, important dates like birthdays and anniversaries, and notable background milestones.

Privacy Concerns and Expert Reactions

The depth of personal data collection has prompted caution from external researchers and ethicists. Observers note that the assistant's focus on mapping real-world human connections pushes the boundaries of typical consumer technology interactions.

Experts point out that users are increasingly willing to plug their entire lives into digital assistants, encompassing emails, calendars, and financial records. This broad access to information significantly increases the scale of what artificial intelligence tools know about individual users.

Meta's Security and Transparency Measures

In response to these findings, Meta emphasizes that Muse is architected with specific user controls. Each user is assigned a dedicated virtual machine that stores user context securely and remains completely inaccessible to other agents and outside users.

Furthermore, Meta states that individuals retain the ability to wipe stored memories or disconnect external services at any moment. The platform is also designed to require human confirmation before executing actions like sending an email or completing a purchase.

Sources

  • WIREDMuse Creates Detailed Profiles of All Your Friends and Family

Continue chronologically

You are readingMeta's Muse AI Agent Creates Detailed Profiles on Contacts
Google Introduces New Model Access Limits for Gemini Users
Older storyGoogle Introduces New Model Access Limits for Gemini UsersOctober 4, 2026 · 3 min

Related entity coverage