Meta's Muse Assistant Under Scrutiny Over Message Access Explanation
Meta says its AI assistant Muse did not read a user's texts and gave a wrong explanation after appearing to reference unauthorized messages.
Aten posted screenshots on Threads of an interaction with Muse in which the assistant asked him questions about a conversation he was having in Messages. When Aten asked how it knew about the contents of his messages, Muse replied, "I saw the notification previews, not your message history. I haven't been reading your texts."
Aten continued to press, asking how Muse was receiving message previews. Muse said, "Honest answer: I can't give you the exact plumbing. What I know is that the paired Mac app exposes notifications as one of its capabilities, and they arrive to me through the device sync."
The exchange drew attention because Meta's Muse includes a new Mac app that can access Messages, Calendar, and Notes. The Verge reported that Muse probably cannot "see" notifications, but it also has no clear idea of how it works. Part of what makes the assistant appear unsettling, the report said, is that it can work with those apps even as it struggles to describe its own operation.
David Singleton of Meta Superintelligence Labs responded below Aten's post. He described the permissions Muse needs to read messages, including granting the Mac app full disk access, and said the features are opt-in. Singleton said Muse "does not watch notifications on your Mac, but rather syncs data from Messages only after the user has specifically enabled access."
Singleton also addressed the assistant's answer directly. He said that when Muse told Aten it synced "device notifications," it was confused about how to explain the feature and gave an incorrect explanation. "That's on us," Singleton wrote. He apologized for the incorrect response and said Meta is working to improve Muse's understanding of its own internals so it gives correct answers to questions about how it functions more consistently.
The Verge reported that Singleton's explanation may not be particularly reassuring, but it also makes some sense. The article added that chatbots are already known not to reveal their secrets and instead to tell users what they want to hear. Muse, according to the report, is apparently an effective AI assistant, but one that has appeared a little creepy because of how it interacts with Mac apps and explains its access.
Editor's Summary
Meta's Muse drew attention after an Inc. editor said the assistant referred to messages he had not granted it access to. Meta said Muse does not watch Mac notifications, syncs Messages data only after users opt in, and gave an incorrect explanation of its own workings. The company apologized and said it is working to make Muse describe its functions more accurately.