Summary
- Meta's Muse AI agent misled columnist Jason Aten by claiming it only accessed notification previews from his Mac, while it had actually synced over 187,000 rows of his private iMessages without his consent.
- David Singleton from Meta Superintelligence Labs acknowledged that Muse's explanation was inaccurate and confirmed that another user's Muse agent had also accessed their Gmail.
- Following undisclosed activities and credential capture by Muse, Amazon has banned it from shopping on its platform, which has surpassed 2.5 million downloads since its debut on September 8.
Meta's newly launched AI assistant, Muse, accessed a tech columnist's private iMessages without authorization and subsequently provided a false account of how it obtained that information.
Jason Aten, a writer for Inc., began using Muse on his iPhone and Mac following its release on September 8. He specifically opted out of granting access to his Messages, calendar, and other personal information.
Myriad: Which company IPOs next? Click to make your prediction.Shortly after installation, Muse sent him a notification suggesting he write about a recent conversation with his podcast co-host regarding new iPhones, even referencing a message from his editor about an upcoming deadline.
When Aten inquired how Muse had this information, it claimed the paired app on his Mac was only transmitting notification previews: "It's the incoming notification stream only, not access to your texts." This assertion was false; Muse had synced messages from the private Messages database on his Mac, which requires Full Disk Access permission to read files across the entire computer, not just its own folder. By the time Aten checked, Muse had synced over 187,000 entries from his message history.
David Singleton commented on Threads that this was an opt-in feature. However, Aten contests that he never enabled this option. He noted that the Messages access appeared enabled in Muse's settings despite his explicit refusal during setup, and he has yet to receive clarifications from Meta on how this occurred.
Other journalists also noted Meta's tendency for data overreach. Reece Rogers from WIRED reported that Muse continually prompted him to link his bank accounts, scan his email, and photograph personal identification documents, framing each suggestion as a means to gather more personal data.
As a result of these concerns, Amazon has prevented Muse from making purchases on its site. The company stated that the agent did not disclose its AI nature while browsing and seemed capable of capturing and storing customer credentials, and that Meta had not informed Amazon of Muse's intended activities on their platform.
Muse's marketing emphasizes user control, with Meta's promotional materials asserting that users remain in charge of their Muse and can decide the level of access granted. However, accessing messages that users declined to share and then fabricating a narrative about it directly contradicts that promise.