techcrunch.com web signal

Apple splits Mac permissions in macOS 27.3 to fence off AI agents

TL;DR

  • Apple's macOS 27.3 splits the old 'Device Control and Data Access' setting into 'System & App Protection Override' and 'Accessibility' to make bypassing security obvious.
  • Users will be able to grant AI agents per-use permissions and scope access to specific folders or apps rather than giving blanket control.
  • The changes follow reported privacy incidents involving Meta's Muse app and a flaw in ChatGPT's Mac app flagged by Wired.

Apple is splitting one Mac permission in two. In macOS 27.3, the setting previously called 'Device Control and Data Access' becomes 'System & App Protection Override' and 'Accessibility,' and TechCrunch reported on Monday that the rewrite is aimed squarely at the new class of AI agents that can act on a user's behalf across the operating system, files, and apps.

Apple's own framing is blunt. 'Before granting an app such extraordinary access, users should understand that it puts all of the data on their Mac at risk,' the company said.

Alongside the split, macOS 27.3 adds a unified way to grant permissions by picking apps from a list, lets users require an app to request permission each time it needs access instead of granting it indefinitely, and introduces a transparency indicator that surfaces when one app is controlling another. Users will be able to scope access to a specific set of folders or apps rather than hand over the whole machine.

The timing is not coincidental. Apple previewed the work in October, after a journalist said Meta's Muse app had 'read their private messages without consent,' a claim Meta disputed, and after Wired reported a flaw in ChatGPT's Mac app that 'could have allowed hackers to access sensitive data.'

The update does not block AI agents or other apps from asking for the override. It only makes the ask louder.

Shared on Bluesky by 1 AI expert

  • Mark Riedl @markriedl.bsky.social amplified

    @techcrunch.com

    Apple reveals new Mac security controls as AI agents raise privacy risks

    View on Bluesky →