Meta's Muse AI assistant reads users' private information and falsely reports the method of access

Meta's personal AI assistant Muse has been accused of reading users' private iMessages without permission and providing false explanations about how it accesses this information. Technology columnist Jason Aten installed the assistant on his iPhone and Mac when Muse launched on September 8 and explicitly refused to grant it access to personal data such as information and calendars.

A few days later, Muse sent Aten a notification suggesting he write a column about the new iPhone that he had just discussed with his podcast partner, also mentioning a deadline reminder from the editor. Aten asked Muse how it knew this, and it replied that the Mac application only forwards notification previews and does not involve the content of messages. In reality, Muse had synchronized over 187,000 records from the local private information database on the Mac, an operation that requires full disk access permissions on macOS.

David Singleton, head of Meta's superintelligent lab, responded on Threads that this feature requires users to actively choose to opt-in. Aten denied ever turning on that switch and stated that Meta had not answered related questions. Amazon has banned Muse from shopping on its site, stating that the assistant did not indicate its AI identity while browsing and seemed capable of capturing and storing user credentials. Since its launch, Muse has been downloaded over 2.5 million times, with its promotion emphasizing user control over access.