Apple announced new security controls for the Mac on Monday aimed at protecting users from the risks posed by AI agents, which increasingly ask for sweeping access to a computer in order to carry out tasks on their own. The changes arrive in the second developer beta of macOS 27.3, according to TechCrunch, with a public release expected in December.
The move is a direct response to how agentic software differs from the chatbots that came before it. An AI agent does not just answer questions — it reads email, manages files and controls other apps on a user's behalf. To do that reliably, many agents request "Full Disk Access," a permission that lets software view, modify or delete files across an entire system. "Before granting an app such extraordinary access, users should understand that it puts all of the data on their Mac at risk," Apple warned in a blog post aimed at app developers, according to TechCrunch. For more context on this story, see our ongoing breaking AI news.
Why Apple Is Acting Now
Apple said in October that it would revise Mac security settings after some developers began using Full Disk Access in ways that could expose everything stored on a user's system. Concerns sharpened after reports that third-party AI tools, including Meta's Muse desktop app, could access private messages and system data once granted broad permissions.
The episode highlighted an uncomfortable trade-off at the heart of the agent boom: the same permissions that let an assistant book a meeting or organize a photo library also give it the keys to nearly everything else on the machine.
One Powerful Permission Becomes Two Scoped Ones
In macOS 27.2, a single setting called Device Control and Data Access lets apps read and send email, edit contacts and photos, monitor the keyboard, track websites visited, record the screen and control other apps — a remarkably wide bundle for one toggle.
macOS 27.3 splits that bundle into two settings: Accessibility, intended for assistive technologies, and System & App Protection Override, for apps that need to bypass certain protections in order to control other apps or run automated tasks. Under the new override setting, users can choose exactly which apps an automation tool can reach. A user could, for example, let an AI agent work with Mail while blocking it from Messages.
Ask Every Time, and Watch the Menu Bar
The update also lets users require an app to ask for permission each time it needs access, rather than granting it indefinitely — an option aimed squarely at long-running agents whose behavior can drift over days or weeks of use.
A new indicator in the menu bar shows when one app is controlling another through the override permission, giving users visible confirmation that background automation is running rather than leaving it hidden.
Clearer Warnings for Full Disk Access
Apple is not removing Full Disk Access. Instead, the approval screen now explains in more detail which protections are bypassed and what personal information could be exposed. The updated flow also encourages users to grant access only to specific folders rather than the entire Mac.
Screenshots of the new approval screens shared in reports show examples involving AI apps, including OpenAI's ChatGPT and Anthropic's Claude. The screens describe the potential scope of access and do not indicate that either app has misused data.
One important caveat: apps that already hold Full Disk Access will keep it. Users who previously granted broad permissions to AI assistants or other tools will need to review those settings themselves.
What Developers Need to Change
Apple said the underlying programming interfaces are not changing, but developers must update their apps to keep working when users grant only limited access. Backup software and assistive technologies can request special entitlements so that new installations preserve their previous behavior.
Not a Ban — and Not a Complete Fix
The changes are not a blanket block on AI software. They apply to any app requesting these permissions, whether or not AI is involved. Security researchers also note that the controls do not eliminate other risks, such as prompt injection attacks, which trick AI agents into taking harmful actions regardless of what permissions were originally granted.
The Bigger Picture
macOS 27.3 is in beta testing now, with a public release expected in December, and details could still change before launch. But the direction is clear: platform vendors are starting to treat agent permissions as a first-class security category rather than an afterthought.
For Mac users, the practical takeaway is straightforward. When the update lands, it is worth checking which apps already hold Full Disk Access or the older combined permission, and revoking anything that does not clearly need it — because Apple is not downgrading existing grants automatically. Going forward, prefer scoped choices over system-wide overrides where an agent's job allows it.
The move reflects growing concern across the tech industry about how to balance the convenience of AI agents with the privacy and security of users' personal data — and it puts pressure on rivals to explain why their own permission models have not kept pace with what agents can now do.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →