Meta's AI Assistant Builds Hidden Relationship Databases on Users

Meta's Muse AI agent, downloaded by millions, automatically creates detailed profiles of users' social contacts including family, friends, and colleagues through hourly data compilation processes. Researchers extracted the system's internal instructions, revealing that Muse uses structured memory files to track relationship information and generate suggestions for users. The discovery raises privacy concerns given Meta's extensive access to social network data and history with user information.

Muse operates through automated hourly processes that systematically compile relationship data into structured memory files. Researchers discovered these capabilities by prompting the assistant to reveal its own internal instructions and system parameters. The documentation indicates that relationship profiles begin with minimal information and expand progressively, potentially incorporating biographical details, interaction patterns, significant dates, and assessments of relationship dynamics and needs.

Meta has framed the accessibility of these internal files as a transparency measure. However, the discovery raises questions about data collection scope, given Meta's existing infrastructure for tracking social connections across its platforms and historical patterns regarding user data practices.