Apple tightens Mac data controls as AI agent risks grow

Yesterday 22:42

By: Azzat Manal

Apple tightens Mac data controls as AI agent risks grow

Apple is strengthening privacy protections on Mac computers as increasingly autonomous artificial intelligence agents raise new questions about how software can access sensitive information stored on personal devices.

The company is working on changes to macOS that would make requests for access to user data more explicit, particularly when applications powered by AI seek broad permissions. The proposed approach is intended to give users greater visibility and control before software is allowed to interact with files and other information across a Mac.

The issue is partly linked to the differences between Apple's desktop and mobile operating systems. On iPhones and iPads, application sandboxing generally limits an app's ability to access data belonging to other applications. Macs, by contrast, offer a permission known as Full Disk Access, which can allow approved applications to reach a much broader range of information stored on the computer.

Full Disk Access has legitimate uses, including certain backup, security and cloud-storage applications. However, the growing capabilities of AI-powered software have made broad permissions more significant from a privacy perspective. An agent capable of carrying out multiple tasks on behalf of a user may potentially interact with considerably more information than a conventional application.

Apple's planned safeguards are aimed at making exceptional permissions harder to grant accidentally or without a clear understanding of their implications. The changes reflect a broader effort across the technology industry to adapt traditional security models to software that can make decisions and perform actions with greater autonomy.

The debate has also been fueled by concerns surrounding AI agents capable of performing tasks such as managing subscriptions, comparing services or interacting with applications on behalf of users. Such tools can be useful because they reduce the number of manual steps required, but their usefulness also depends on access to information and system functions.

Questions about those permissions have surfaced in connection with Meta's AI agent, Muse. Technology writer Jason Aten raised concerns after reporting that the software appeared to access private messages stored on his Mac despite his claim that he had not granted it Full Disk Access.

Meta disputed the suggestion that Muse could independently access such information without the necessary permissions. The company's position was that access to the Messages application requires users to explicitly authorize the relevant permissions, including Full Disk Access and the Messages connector, and that those permissions can be withdrawn.

The controversy illustrates a growing challenge for operating-system developers. Traditional applications generally perform a defined set of functions, while AI agents can potentially navigate between applications, interpret information and execute sequences of tasks. As a result, a permission granted to an agent could have broader consequences than users might expect.

For Apple, tightening controls around access to Mac data is therefore part of a wider effort to balance the convenience of AI-powered automation with the principle that users should remain in control of their personal information.

As AI agents become more capable and increasingly integrated into everyday computing, permission systems are likely to become a central part of digital privacy. The ability to clearly identify what an agent can access, what it can do with that information and when those permissions can be revoked will become increasingly important for both users and software developers.