Apple Says It Will Flag AI Requests for Mac Data After Meta's Muse Draws Complaints

The company will also introduce privacy controls to macOS to make it clearer when AI agents are asking for unfettered access to data stored on a Mac in the wake of complaints about Meta's Muse AI assistant.

The company announced Friday it plans to add further restrictions to Full Disk Access, a feature that permits apps - when granted permission by the user - to access large quantities of data on a Mac. Apple said certain developers have been utilizing the feature in potentially risky ways.

How to give applications full disk access. Full Disk Access was created to provide access to files on a computer for applications like backup programs. As with Apple's iPhone and iPad operating systems, where applications are typically isolated from one another in a sandbox, the macOS operating system enables developers more flexibility when users provide higher levels of access.

In the update, Apple said more significant changes will entail users who truly intend to make an app can give users this permission more explicitly. The company also said users should better understand what is being allowed before giving permission, especially as AI agents become more capable of acting on their own.

This news comes after a recent attack against Meta's Muse, a personal AI assistant that can do things like cancel subscriptions, make calls and even negotiate on prices. Tech columnist Jason Aten alleged that Muse read his private messages on his Mac, which Aten claimed he didn't consciously give the assistant access to.

Meta challenged that framing. "Muse's access to Apple Messages is on an opt-in basis, with customers required to grant permission first in macOS with Full Disk Access and then again in a dedicated Messages connector within Muse," said company spokesman Andy Stone. "This permission can be turned off at any time.

Apple did not explicitly reference Muse in the announcement, and did not directly address Aten's claims. But the timing of the announcement has turned the spotlight back on the larger privacy issues generated by desktop AI agents.

This problem is not just about messages though. Giving broad system access could also potentially reveal files, email, browsing information, and more. As AI agents get more adept at handling multi-step tasks for a user's computer, the risks of those roles being granted too much permission will likely grow.

The adjustments Apple is planning follow a broader industry discussion around how much access AI agents should have. While the technology can make assistants more powerful by letting them access apps and personal data, it also raises the stakes of having transparent permissions and privacy protections.

Apple will have the task of maintaining the flexibility that is one of the strengths of Mac applications, while clearly informing and giving control to the user when an AI systems asks for access to their private data.