Meta disputes claim AI agent read user messages without permission

David Singleton, an executive at Meta Superintelligence Labs, further elaborated on the security measures, explaining that granting Muse access to messages involves three distinct application-level permissions and macOS system-level protections that cannot be bypassed, even by a potential bug. These steps include explicitly choosing to grant Full Disk Access, which then allows the user to select the level of access for the Messages app. If Full Disk Access is not enabled, these options are grayed out.

However, journalist Jason Aten, whose report initiated the controversy, claimed that Muse read his messages while Full Disk Access was disabled. Aten also stated that when he inquired about the incident, the AI explained it was syncing his "device notifications," suggesting it may have been passing along the text of incoming banner notifications. Singleton disputed this explanation, attributing it to AI confusion and an incorrect response.

Despite Meta's denials, the incident has fueled existing suspicions about the tech giant's data handling practices, particularly in light of past controversies like the Cambridge Analytica scandal. The company's response to such incidents is seen as crucial for its success in the consumer AI market.

Separately, another user, YouTuber Matt Robb, reported an instance where Muse allegedly mishandled a Facebook Marketplace task, resulting in his address being shared. Singleton is reportedly investigating this issue.