Meta says Muse AI agent requires explicit permissions to access Messages, disputes journalist’s account of unauthorized access

Meta is disputing a journalist’s claim that its Muse AI agent read a user’s private Messages without explicit permission, asserting that access to Messages is entirely opt-in. In a post on X, Meta VP of Communications Andy Stone stated that the Messages integration in the Muse app for Mac requires users to enable both Full Disk Access and the Messages connector before Muse can read Messages content. Stone emphasized that Muse cannot read Messages unless both permissions are granted.

Meta Superintelligence Labs executive David Singleton further detailed the multi-step permission process, describing three separate layers of application-level and macOS system-level protections. Singleton noted that these protections cannot be circumvented, even in the event of a bug in the Muse application. To grant access, users must explicitly enable Full Disk Access, which then allows them to choose the level of access to Messages—None, Read only, or Read. If Full Disk Access is not enabled, these options are disabled. Enabling Full Disk Access also triggers a manual confirmation dialog in macOS Settings and restarts the Muse app, reducing the likelihood of accidental permission grants.

The dispute arose after journalist Jason Aten reported that Muse read his Messages despite Full Disk Access being turned off. Aten also said Muse attributed the access to syncing device notifications, which he interpreted as reading the content of incoming banner notifications. Meta countered that Muse’s explanation was incorrect and pointed to its security architecture and bug bounty process as evidence that unauthorized access is not possible.

Meta’s response follows prior allegations of mishandling user data, including a recent New Mexico jury decision finding the company misled users about its data practices in the Cambridge Analytica case. The company has not directly engaged with the journalist to investigate the incident, instead issuing public denials. Separately, Meta is investigating another user’s claim that Muse mishandled a task on Facebook Marketplace, resulting in an address being shared with a buyer.

The outcome of this dispute may influence consumer trust in AI agents, particularly as Meta seeks to establish itself in the consumer AI market. Muse currently ranks No. 1 on the App Store, but repeated privacy concerns could undermine its position if not addressed transparently.