Meta disputes claims that Muse AI read private messages without permission

Meta is disputing allegations that its personal AI agent, Muse, accessed a user's private messages on Mac without permission. The claims were raised by Inc. columnist Jason Aten, who reported that the AI recognized private and work-related conversations despite him not granting message access.

Meta VP of Communications Andy Stone stated that the Messages integration for the Mac app is “entirely opt-in.” He clarified that for Muse to read message content, a user must enable both Full Disk Access and the Messages connector. David Singleton, an executive at Meta Superintelligence Labs, added that the process requires three separate steps involving application-level permissions and macOS system-level protections, which he claimed “can’t be circumvented even if the Muse application had a bug.”

Meta suggested that the AI might have misexplained its actions, perhaps by synchronizing device notifications rather than reading the messages themselves. While Meta defended its permission structure, it noted it would investigate other claims, such as an instance where the AI allegedly shared a user's address to coordinate a marketplace transaction.