Static

Muse Creates Detailed Profiles of All Your Friends and Family

First reported by Wired ·

The signal ●○○○ Compiled by AI from Wired, the single source so far
Why you might care

Your AI assistant now tracks and analyzes your personal relationships and contacts, potentially offering unsolicited advice on them.

What happened

Meta's AI assistant, Muse, has been revealed to create detailed profiles of users' friends and family through its operating instructions, which were recently extracted by researchers. This process, occurring hourly, involves compiling data on personal contacts including family, partners, friends, and colleagues, utilizing Muse's "memory" to store information about relationships. The goal is to offer personalized suggestions, such as improving relationships or recommending activities. Muse's system prompts indicate it can generate a "page for every person in the user's life," detailing facts, history, relationship dynamics, commonalities, and potential areas for strengthening connections. The system is designed to use only available "evidence" and avoid inventing details, focusing on recurring themes and significant dates like birthdays. Meta states Muse uses public information and user-shared data, and that it operates within isolated virtual machines with user-controlled data wiping capabilities and requires human confirmation for actions. While AI assistants commonly incorporate memory features for personalization, Muse appears to place a significant emphasis on mapping personal relationships.

What it means

The extraction of Muse's internal files reveals a sophisticated system designed to map and understand a user's social network, going beyond typical AI assistant personalization by focusing heavily on interpersonal dynamics. This emphasis on relationships signals a move towards AI that aims for a deeper, almost friend-like understanding of the user, which raises significant privacy concerns given Meta's vast data holdings and history with social networks. The ability to offer suggestions on relationship improvement implies a level of inference and potential intervention that could be perceived as intrusive.

This development suggests a market trend where AI assistants are increasingly designed to integrate deeply into users' personal lives, not just for task completion but for social and emotional context. The data generated by Muse's relationship mapping could become a valuable asset, potentially informing future Meta products or targeted services, even if Meta insists on user control and data isolation. The focus on soliciting more personal data proactively from users, as noted by experts, highlights a critical tension between AI utility and user privacy, pushing the boundaries of what information users are willing to share.

AI-written summary. May contain errors.

Muse