Apple is tightening macOS Full Disk Access to limit AI agent risks

Apple is changing how Mac software gets permission to access everything on a hard drive, warning that the rise of powerful, autonomous AI agents has dramatically increased the risk of data theft.

In an update posted to its developer site, the company said it will introduce new controls to ensure that macOS users can only grant "Full Disk Access" through "very explicit user action." The setting, which was originally designed to let backup software run smoothly, bypasses Apple's standard privacy protections.

Why is Apple changing macOS Full Disk Access?

Apple warned that some developers are now using the permission in ways that put users at risk, exposing emails, messages, browsing histories and files.

"As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially," the company wrote. Apple added that when these apps handle communications, the exposure can also compromise the privacy of the people the user is talking to.

Currently, macOS users can grant Full Disk Access by navigating through System Settings to the Privacy & Security menu, then manually adding and enabling an app. Once granted, the permission allows an application to bypass normal boundaries, reading everything from local chat histories to Time Machine backups.

How does the Meta Muse controversy tie in?

The policy change comes shortly after a controversy involving Meta's Muse app. Jason Aten, a columnist for Inc., reported that the AI chatbot was able to read his private messages on macOS even though he did not recall giving it permission.

Meta disputed the claim. A spokesperson for the company said that the software can only read messages if a user explicitly chooses to opt in, which requires enabling both Full Disk Access and a specific Messages connector.

However, the incident highlighted how easily users can grant sweeping access to desktop-based AI tools without fully realising the consequences. Security concerns around desktop AI have been mounting elsewhere, too; a vulnerability discovered in OpenAI’s ChatGPT app for Mac recently showed how hackers could potentially exploit the tool to access sensitive user files.

What happens next for Mac users?

Apple has not yet specified when these tighter controls will roll out, or which version of macOS will be the first to include them.

For businesses managing fleets of Macs, the company's platform deployment guide notes that organisations can still manage these settings through device management payloads, which will apply the most restrictive settings when multiple policies conflict. For everyday users, however, the era of clicking through permissions without thinking is about to get a lot harder.