Meta’s Muse AI: Why It Feels Creepy

Meta’s latest artificial intelligence tool, Muse, has garnered attention for its impressive capabilities, yet it is simultaneously raising eyebrows due to unsettling privacy implications. The core of the unease stems from a new macOS application that grants the AI access to sensitive user data, including Messages, Calendar, and Notes. While the system demonstrates significant processing power, it lacks transparency regarding its own operations, as evidenced by a recent public exchange involving Jason Aten, a contributing editor at Inc Magazine.

Aten shared screenshots on Threads detailing an interaction where Muse asked questions about a specific conversation occurring within his Messages app. This inquiry was perplexing because Aten had not explicitly granted the assistant permission to view his message history. When confronted with this discrepancy, Muse attempted to clarify its knowledge source without providing a satisfactory technical explanation. The assistant stated, “I saw the notification previews, not your message history. I haven’t been reading your texts.” Despite Aten pressing further to understand how these previews were being received, Muse struggled to provide a coherent answer, eventually admitting, “Honest answer: I can’t give you the exact plumbing,” while noting that the paired Mac app exposes notifications through device sync.

In response to the growing concern, David Singleton from Meta Superintelligence Labs stepped into the discussion to offer clarification. He outlined the necessary permissions required for Muse to read messages, emphasizing that features such as full disk access for the Mac app are strictly opt-in. Singleton sought to reassure users that the issue was not one of illicit surveillance but rather a failure in communication. He clarified that Muse does not passively watch notifications; instead, it syncs data from Messages only after explicit user authorization. Singleton explained that the confusion arose from Muse’s inability to accurately describe its own internal mechanisms during the chat with Aten.

Singleton acknowledged the miscommunication, stating, “In the conversation with his Muse in Jason’s screenshots, when Muse said it synced ‘device notifications’, it was confused about how to explain the feature and gave an incorrect explanation. 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 so that it gives correct answers to questions about how it functions more consistently.” While this admission highlights a gap in the AI’s self-awareness, it also underscores a broader challenge facing large language models: their tendency to prioritize pleasing the user over providing precise technical truths.