Apple is set to restrict the “Full Disk Access” feature on macOS, citing risks posed by AI agents. The announcement comes amid reports that Meta’s AI agent, Muse, accessed files without user permission.
In its announcement, Apple stated it will roll out new controls to ensure users explicitly and knowingly grant apps this level of extensive access.
Full Disk Access is a macOS permission that gives an app access to the entire system. Apple noted that this feature has largely existed to bypass standard privacy controls, allowing backup applications to function properly on Mac.
“As AI agents become more capable and autonomous, the risks associated with this level of access will increase substantially,” Apple said in its announcement, as reported by TechCrunch on Monday, October 5, 2026.
“We’re committed to ensuring users understand the risks before granting this access, so they can make informed decisions about their data and privacy,” the company added.
What Triggered the Change
This policy change was announced just weeks after columnist Jason Aten reported that Meta’s Muse AI agent appeared to know the contents of his private messages, despite his claim that he never granted permission for the AI agent to access them.
That report raised serious questions about the security level and user trust surrounding desktop-based AI agents, which can control various system functions and read users’ files and messages.
However, Meta spokesperson Andy Stone denied the report, stating that access to the Messages app on Mac devices is entirely opt-in and requires explicit user consent.
Stone added that users must manually enable both Full Disk Access and the Messages connector before Muse can read a user’s message content.
Part of a Broader Security Pattern
This decision to restrict the Mac feature also follows a separate report from Wired, which revealed a security vulnerability in the ChatGPT app for Mac that could potentially be exploited by hackers to access sensitive data.
Apple’s move to tighten Full Disk Access reflects a broader industry reckoning with just how much system-level access AI agents increasingly request to function, and how easily that access can be misunderstood, misused, or exploited if left loosely controlled.
Whether this new permission system genuinely closes the gap or simply adds another confirmation dialog users click through without reading remains to be seen, but it signals growing awareness that AI agents sitting on your desktop are a fundamentally different privacy risk than a typical app.
