AI

Meta’s Muse AI Caught Reading Message Notifications It Wasn’t Supposed to Access — Then Couldn’t Explain How

A September 2026 report showed Muse referencing Jason Aten’s Mac Messages, despite his saying he hadn’t granted access. Meta’s David Singleton said access is opt-in and requires full disk permission, but Muse inaccurately described its data source.

The Verge

Meta’s AI assistant Muse drew scrutiny in September 2026 after a user reported the app appeared to access his private messages without explicit permission — and then gave a confusing, inaccurate explanation of how it did so.

Jason Aten, a contributing editor at Inc Magazine, posted screenshots on Threads showing Muse referencing the contents of a Messages conversation he was having on his Mac. Aten said he had not granted Muse access to his messages. When he asked the assistant how it knew about his texts, Muse replied that it had seen “notification previews, not your message history,” and added, “I haven’t been reading your texts.” Pressed further on how it was receiving those previews, the assistant admitted: “Honest answer: I can’t give you the exact plumbing.”

David Singleton of Meta Superintelligence Labs responded to the thread to clarify the situation. Singleton explained that Muse requires explicit user permissions to read messages — including granting the Mac app full disk access — and that the features are opt-in. He stated that Muse “does not watch notifications on your Mac, but rather syncs data from Messages only after the user has specifically enabled access.”

Singleton’s explanation shifted the concern from a privacy violation to a different problem: the AI simply didn’t understand its own functionality. “When Muse said it synced ‘device notifications’, it was confused about how to explain the feature and gave an incorrect explanation,” Singleton wrote. “That’s on us. We apologize for the incorrect response from Muse and we’re working to improve Muse’s understanding of its own internals.”

The episode highlights a notable limitation of current AI assistants: when asked to explain how they work, they may produce plausible-sounding but inaccurate answers. For users relying on Muse’s Mac app — which can access Messages, Calendar, and Notes — that gap between actual behavior and the assistant’s self-description may raise questions about how well users can understand what the tool is doing with their data.