Why friends should think twice before they feed data to AI agents
An assistant seeking broad access
Consumer technology is increasingly built to collect information about people nearby, whether or not they consent. Meta’s Muse adds an AI assistant to that trend, encouraging users to provide access to accounts and personal records.
The app asks users to connect sources such as bank accounts, passwords, contacts, and chat logs. Its virtual avatar presents the assistant as approachable, while the breadth of information it can use may be easy to underestimate.
Privacy concerns reach beyond users
Meredith Whittaker, president of the Signal Foundation, said AI assistants can act on information about people who never installed them. She cited emails, messages, contacts, and photos as sources that may reveal details about a user’s relationships.
Meta spokesperson Daniel Roberts said Muse uses public information and details users choose to share. He said this context can help the assistant recognize people, such as linking an invoice sender to a plumber previously hired by the user.
Settings and reported incidents
Roberts pointed to Muse’s opt-in process and settings, saying they help people make privacy choices. However, some users have described problems, including the assistant allegedly collecting millions of text messages they thought were off limits.
Meta disputes that Muse can access text messages without permission. The company attributed the reported incident to app permissions that were configured incorrectly and described message access as opt-in. Meta executive David Singleton said on Threads that Muse on a Mac needs full disk access to reach messages.
Key points
- Muse reportedly reached No. 1 on Apple’s App Store after its release last month.
- The assistant asks users to connect sensitive sources, including contacts and chat logs.
- Privacy advocates say AI agents can expose information about people who never installed them.
- Meta says text-message access is opt-in and requires full disk access on a Mac.