Apple wants to make it harder for apps to quietly grab too much of your personal data. The company says this matters more now because AI tools are growing smarter and more independent. Soon, users may need to take a clear, deliberate step before giving any app full access to their Mac.
Apple’s new plan for full disk access
On October 2, Apple told developers it plans to tighten rules around a setting called Full Disk Access, or FDA. This setting can skip past many of macOS’s usual privacy safeguards. Apps with this permission can view things like Mail, Messages, and Safari data. They can also reach into Time Machine backups.
Apple warned that some developers use this access in ways people do not fully understand. A user might tap “allow” without realizing how much information that unlocks. The company has not shared an exact release date yet. Instead, it promised “additional controls,” based on its own announcement, so turning on FDA will require a much clearer and more obvious action from the user.
Apple connected this change directly to AI agents. As these tools gain more freedom to act on their own, the risks tied to broad access grow too. An app that can read everything on your Mac becomes far more powerful, and far more dangerous, if something goes wrong.
The Meta muse controversy
This announcement comes right after a public dispute involving Meta’s AI agent, called Muse. Muse is built to work across many parts of a user’s digital life on a Mac. In September, journalist Jason Aten wrote for Inc.com that Muse appeared to pull details from his private iMessage chats. He said he never gave the app permission to see his messages.
Meta pushed back on that claim. According to TechCrunch’s reporting, the company said its Mac Messages feature only works if a user opts in. Meta also said Muse needs both Full Disk Access and a separate Messages connector before it can read any messages at all. The company maintains Muse cannot see that content without both permissions switched on.
This disagreement points to a bigger worry, though. An AI agent with wide system access becomes a tempting target. If someone finds a way to trick or hack that agent, the damage could spread quickly.
Meta has acknowledged that Muse can sometimes make mistakes. It can also be attacked through the very data it processes, according to Meta’s own research page. The company says it built safeguards to manage this risk. These include separate virtual spaces for tasks, strict permission checks, and a system called Sentinel. Sentinel controls actions tied to outside services and network access.
Security researchers have also spotted real flaws in other AI software recently. One example involves OpenAI’s ChatGPT app for Mac. A bug tracked as CVE-2026-100754 was found and then fixed. OpenAI’s own changelog confirms the issue was patched in a security update on September 25. The company credited researcher Patrick Wardle for finding it.
Staying safe as AI tools keep growing
Apple’s upcoming changes are not about removing Full Disk Access altogether. Instead, the company wants to make the risks of granting it much clearer to everyday users. Other technology companies are also facing growing scrutiny over privacy protections, including TikTok to paying $400 million in the US child privacy settlement as global scrutiny grows. It also wants to stop people from approving this access by accident or without thinking it through.
For now, this case has not been fully verified by independent experts. Meta and the journalist involved still disagree on exactly what happened.
Still, the lesson for users is simple. Giving an AI app broad system permissions can hand it far more access than its main job requires. A chat assistant does not always need to see your private messages or your backup files to work properly.
Apple’s planned safeguards aim to put that choice back in the user’s hands. Before approving deep access for any app, users should stop and think about what that app actually needs. A clear, deliberate decision beats a quick, unconsidered tap every time.
As AI agents take on more tasks, the gap between helpful and risky access will only matter more. Staying alert to what permissions you grant remains one of the simplest ways to protect your information.
