Apple is implementing stricter controls on Mac systems to restrict how AI agents access user data, following concerns that developers may be using full disk permissions in ways that compromise user privacy. The company announced the changes on its developer blog, revealing that the latest macOS beta now requires users to take highly explicit actions to grant applications like ChatGPT, Claude, and Gemini this level of access.
The issue gained attention after Inc. columnist Jason Aten discovered in September that Meta’s Muse Mac app could access his private messages, despite his belief that he had denied the permission. While Meta maintained that the feature was opt-in and argued that Aten had inadvertently granted access, the incident prompted Apple to intervene. Apple noted that some developers were exposing sensitive information—including files, emails, messages, and browsing history—without users’ full knowledge or understanding.
Under the new system, the Full Disk Access setting in macOS now triggers a prominent warning screen when users attempt to grant permissions. According to screenshots shared by X user @madeby_ldt, the pop-up warns that the Mac will no longer be able to protect data from the specified AI app. Users must review and explicitly acknowledge that the application will be able to read, change, delete, and export all data, view hidden and recently deleted photos, and monitor browsing history.
Although these controls do not prevent users from granting access, they are designed to ensure that individuals pause and consider the risks. The update is particularly relevant for Mac Studio models, which Apple markets for running local AI models and agents. The company stated that as AI agents become more autonomous, the risks associated with such extensive access will increase, and it is committed to helping users make informed decisions about their privacy. The feature is currently available in macOS 27.3 beta 2, with no confirmed timeline for a public release to all users.
Apple finally woke up. This is exactly what we need for autonomous agents moving forward.
Funny how ‘full disk access’ sounds normal but actually means they can delete everything instantly.
Hope this makes it into the stable release quickly. Beta users are brave enough for now.
It’s not about preventing access, it’s about informed consent. Much better than hidden permissions.
I will definitely be clicking deny every time. I don’t want an AI watching my browser history.
Jason Aten’s story scared me. Glad Apple is actually listening and taking concrete protective steps now.
Skeptic here: will big tech just lobby Apple to bypass these warnings in the final release?
Does this apply to local models running on Mac Studio too, or just cloud-based agents?
This is the right move. Stop trusting developers to do the right thing with sensitive data.
Gladly. I was terrified my AI assistant could read my banking logs without me knowing anything.