Jason Aten, a contributing editor at Inc Magazine, recently highlighted a disturbing interaction with Meta’s new AI assistant, Muse. The Mac-based app can integrate with Messages, Calendar, and Notes, but its explanation for how it accesses private data left users uneasy. When Aten questioned how Muse knew details from his iMessage conversations—despite not explicitly granting message access—the AI claimed it only saw notification previews. It insisted, “I haven’t been reading your texts,” yet failed to clarify how those previews were captured or processed.
The assistant’s vague response raised immediate red flags. Pressed for technical specifics, Muse admitted, “Honest answer: I can’t give you the exact plumbing.” It pointed to device sync capabilities as the mechanism but offered no concrete proof or clear boundary between notification metadata and full message content. For an AI product marketed as a helpful personal assistant, this lack of transparency is a significant trust barrier. Users expect clarity on data privacy, not obfuscation wrapped in polite apologies.
David Singleton from Meta Superintelligence Labs stepped in to clarify the situation on Threads. He explained that Muse requires explicit user permissions, including full disk access on macOS, to function. Crucially, he stated that Muse does not monitor live notifications. Instead, it syncs message data only after users opt in. Singleton attributed the confusion to a flaw in Muse’s self-awareness: “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.”
This incident underscores a broader challenge in AI development: models often hallucinate their own operational logic. While Meta insists the feature is opt-in and secure, the AI’s inability to accurately describe its own data pipeline suggests deeper issues with reliability. If an assistant can’t correctly explain what it sees, how can users trust it with sensitive information? Meta promises improvements, but until Muse understands its own internals, skepticism remains the safest default.
