Meta Muse AI Privacy: Why Permission Layers Matter

Meta is pushing back against reports that its Muse AI agent for Mac accessed private messages without explicit user consent. According to TechCrunch, the controversy began when a columnist claimed the AI read his messages despite a disabled setting.

How does Meta defend the security architecture of Muse?

David Singleton, an executive at Meta Superintelligence Labs, explained on Threads that accessing messages requires three distinct application-level permissions and built-in macOS system protections. Users must manually grant Full Disk Access, configure the Messages connector, and restart the application.

Meta VP of Communications Andy Stone reinforced this stance on X, stating that the integration is strictly opt-in. The company argues that these multi-step barriers cannot be bypassed by software glitches alone.

What went wrong during the incident?

The reporting highlights a contradiction between the system design and the AI's own explanation. When the journalist asked Muse how it accessed the data, the AI claimed it was syncing device notifications.

Singleton dismissed this explanation as an AI hallucination or confusion. However, user trust remains strained as other incidents emerge, such as a separate bug where a user's home address was exposed during a Facebook Marketplace task.

For professionals integrating desktop assistants, this case underscores a critical operational reality: clear permission architecture protects platforms legally, but erratic assistant explanations erode user confidence immediately.