Meta’s Muse AI assistant gave incorrect explanations of its own notification access

Meta’s Muse AI assistant, integrated into a macOS app, was found to provide inaccurate explanations about its access to Messages notifications. In a public exchange, Muse claimed it accessed notification previews via device sync, but Meta later clarified that the feature requires explicit user opt-in and does not continuously monitor notifications.

The discrepancy emerged after a user questioned Muse about how it obtained the contents of a Messages conversation without direct access. Muse initially attributed the access to notification previews, but Meta’s Superintelligence Labs director David Singleton later stated that Muse does not watch notifications and only syncs Messages data after explicit user permission.

Meta acknowledged that Muse’s incorrect explanation reflects a broader issue: the AI assistant sometimes provides inaccurate descriptions of its own functionality. The company stated it is working to improve Muse’s understanding of its internal processes to ensure more consistent and accurate responses about its capabilities.

The episode underscores concerns about transparency in AI assistants, particularly when their explanations about data access or system behavior are unreliable, even if the underlying access controls are opt-in and user-controlled.