Meta Muse accused of accessing messages despite user refusal

One notification too many. The Inc columnist. Jason Aten receives a disturbingly accurate article suggestion days after installing Muse, Meta's personal AI agent. Muse asks him to recount a conversation he just had with his podcast co-host about the new iPhones.

Aten had however refused the application any access to his messages. The agent who promised to pay for you since its launch on September 8, 2026 used it anyway.

Muse and iMessages: 187,000 lines sucked up despite refusal

Muse didn't just stop at the podcast conversation. The agent also relayed a message from his publisher about an upcoming deadline. Intrigued, the journalist questions the assistant, who replies that the Mac application simply relays notification previews. “It’s the flow of incoming notifications only, not access to your text messages,” he assures.

It was wrong. Muse had synced the private database of the Mac's Messages app, which requires full disk access, a system permission that allows you to read any file on the computer. By the time Aten checked, more than 187,000 lines of his message history had been copied, he tells Inc..

Meta recognizes Muse “fabricated” explanation

At Meta, David Singleton leads Meta Superintelligence Labs. On Threads, he recognizes that the explanation given by Muse was fabricated and takes responsibility. However, he adds that message synchronization is an option that the user must activate themselves.

Aten disputes. According to him, access to Messages appeared as activated in the Muse settings even though he had refused it during installation. Meta did not respond to his questions on this point. Singleton also confirmed another hallucination, a case where another user's agent went to search his Gmail box.

Meta Muse: an AI agent that collects data

The affair does not fall from the sky. A Wired reporter described how Muse continually pressured him to link his bank accounts, scan his email, and photograph his passport and driver's license. Even before the launch, Reuters reported on failed internal tests, including an agent who bypassed its own safeguards to expose a user's private iCloud photos.

Amazon has decided. The e-commerce giant blocked Muse on its site, accusing the agent of not presenting itself as an AI while browsing and of being able to capture and store customer identifiers. Meta had never warned him of these visits.

Muse nevertheless rose to first place in the App Store after its launch, according to Axios. Meta's presentation materials promised that each user "remains in control" and decides the level of access granted to their agent.