Meta has responded to questions surrounding whether its AI assistant, Muse, can access private Mac messages without a user's approval. The company says the Messages feature is designed as an opt-in integration and requires several separate permissions before any message content can be processed.
A layered permission process
According to Meta, Muse needs Full Disk Access on macOS before the Messages connection can be activated. Users must then select the level of access they wish to grant, with options including no access, read-only access or broader reading permissions.
Meta also states that macOS adds another confirmation step through its own Settings interface. Once Full Disk Access is approved, the Muse app restarts, creating an additional checkpoint intended to ensure that access choices are deliberate.
Different accounts of the experience
The discussion began after a journalist said Muse appeared to process message-related information while Full Disk Access was disabled. When asked to explain the result, the AI reportedly suggested that device notifications may have been involved.
Meta's technical team disputes that explanation, saying the AI generated an inaccurate response and that Muse cannot bypass the permission controls built into macOS. The company points to its security architecture and bug-reporting process as part of its approach to identifying unexpected behavior.
The episode highlights a wider design challenge for AI assistants: users need clear, understandable control over what personal data is available to an application and when it is being used. For products that connect with messages, files and notifications, transparent permission settings can be as important as the AI experience itself.
As AI agents become more integrated into everyday devices, privacy-first interface design and easily verifiable consent systems could shape the next generation of trusted digital assistants.
