Meta's Muse creates detailed profiles of users' contacts
Researchers Karan Joshi extracted Muse internal files via chat interface, revealing that the Meta AI assistant can generate a dedicated page for every person in a user's life. The files show that Muse compiles structured text files—called memory—about family, friends, colleagues, and people the user follows. The pages include sections such as Facts, History, Relationship, In common, Open…
Key points
- Researchers extracted Muse internal files via chat interface, revealing its ability to build a page for each person in a user's life.
- Muse stores data in a user‑specific virtual machine and seeks human confirmation before actions like sending email.
- Meta says Muse logs all activity and limits invented details, but privacy experts warn about the breadth of data collected.
Meta says each user has a separate virtual machine that stores data and can be wiped or disconnected. The assistant seeks human confirmation before sending emails or making purchases and logs all activity. Researchers note that Muse’s focus on personal relationships is more intense than other assistants, raising privacy concerns. Carissa Véliz and Miranda Bogen warn that the breadth of data collected could lead to a ballooning of what the system knows about users.
The disclosure highlights how AI assistants can gather and structure personal information, prompting questions about transparency and user control. Meta maintains that the files were made available for transparency, but the extent of data Muse can collect remains a point of debate.
Model page: Muse →
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Instinct hits $10B valuation as text-based AI agents expand · 1 src
- AI agents use 5x more tokens than humans, per OpenRouter data · 1 src
- SGD vs. Adam: How Machine Learning Optimizers Actually Learn · 1 src
- GitHub makes new dashboard the default for all users · 1 src
- OpenAI launches Dots, always-on AI agents in ChatGPT · 65 src
Comments
via GitHub Discussions