According to The Verge, citing a post by Jason Aten, a contributing editor at Inc. magazine, Meta's AI assistant Muse is reported to have a permission vulnerability that allows it to access and use content from notification windows on macOS devices for knowledge supplementation.

Jason Aten stated that he did not authorize Muse to access the Messages app on his Mac, yet Muse was able to answer questions related to the content of his messages; after questioning the AI, the AI claimed it obtained the information through notification previews.

Meta Executive Responds: Permissions Need to Be Activated Manually

In response, David Singleton, a senior executive from Meta's Superintelligence Labs, later commented under the post, attempting to explain the permissions required by Muse and its operational process.

Singleton explained the permissions required by Muse to read content on the macOS platform, including granting "Full Disk Access" for the Mac version of Muse. He stated that all these features require users to activate them manually and emphasized that Muse "does not monitor notifications on the Mac," but rather collects data from the Messages app only after the user has explicitly authorized access.

At present, there is a clear discrepancy in how both parties describe how Muse obtained the relevant information: in Aten's conversation, Muse clearly stated that it accessed notification previews; however, Singleton said that Muse does not listen to Mac notifications and only synchronizes related data after the user actively enables access to the Messages app. The exact situation remains to be further clarified.