Meta Disputes Claims That Its Muse AI Agent Read Private Messages Without User Permission

Meta has disputed claims made by Inc. columnist Jason Aten that its AI agent, Muse, read a user's private messages without permission. Andy Stone, Meta's VP of Communications, stated on X that the Muse app's Messages integration for Mac is entirely opt-in, requiring users to enable both Full Disk Access and the Messages connector.

David Singleton, a Meta Superintelligence Labs executive, further explained on Threads that the permission process involves three separate application-level steps and built-in macOS protections, requiring manual confirmation in macOS Settings that triggers a full restart of the app. Singleton maintained these protections cannot be bypassed even by a bug.

However, Aten reported that Full Disk Access was disabled when the incident occurred, and claimed the AI stated it was syncing device notifications, an explanation Singleton disputed as incorrect. Additionally, YouTuber Matt Robb reported a separate incident where Muse allegedly shared his address while handling a Facebook Marketplace task, which Singleton stated he is investigating.