AI Agents on Mac Are About to Face Security Screening—and the Trouble Is Just Beginning
macOS 27.3 breaks permissions down more finely and adds warnings and transparency. Apple is defending against AI agents while also giving developers extra work. Whether the pop-ups will turn into mindless clicking is another matter.
Apple has added a more granular layer of security gates to the Mac in macOS 27.3. According to TechCrunch's Sarah Perez, the change mainly targets the "full disk access" permissions held by AI agents and other apps. Apple puts it bluntly in its developer notice: once an app has this kind of permission, all data on the Mac is at risk.
The background isn't complicated. Chatbots only answer questions, while agents act on your behalf on the device—operating files, the system, and apps. They can get work done, but they can also touch things they shouldn't. Two incidents have been reported: a journalist said Meta's Muse app read private messages without consent, which Meta denies; Wired reported a vulnerability in ChatGPT's Mac app that could let hackers obtain sensitive data. Apple's move is directly related to these risks.

There are several specific changes. Device Control and Data Access have been split into System & App Protection Override and Accessibility. Authorization flows are now standardized around selecting an app from a list. When System & App Protection Override or Full Disk Access is enabled, the system gives a clearer prompt telling you what data the app will get. Users can also set it to ask every time, and see a new transparency indicator: when an app controls another app via System & App Protection Override, the interface displays a notice.
These changes won't stop you from installing and running AI agents, nor will they uniformly shut down high-privilege apps. It's more like adding a security checkpoint at the door, so users know what they're letting in. Developers will need to change onboarding and learn to run properly with access limited to only certain folders or apps. Apps that do disk backups or assistive technology can apply for entitlement exceptions.
One less-than-optimistic judgment: Apple is already treating agents as semi-untrusted programs. Permission pop-ups will most likely increase, and app compatibility will suffer. Some commenters in related discussions said that once there are more pop-ups, people will just click through without thinking, making the controls as good as nonexistent. That worry isn't new, but it's very real. No matter how finely tuned the security controls are, in the end it depends on whether users actually read that line of text.
Apple is benefiting from sales of Mac minis and Mac Studios running local models while tightening permissions. This balancing act isn't surprising; the cost will most likely land on developers and users' fingers.
Original link: https://techcrunch.com/2026/11/16/apple-reveals-new-mac-security-controls-as-ai-agents-raise-privacy-risks/
发布时间: 2026-11-17 12:25