Apple Overhauls Mac Security in macOS 27.3 to Address AI Agents With Full Disk Access
Apple is reworking Mac security for AI agents: programs that work with files and other apps on the user's behalf. Many of them currently ask for "Full Disk Access," which exposes all the data on the computer. In macOS 27.3, the settings will be split into two categories, with risk prompts and an "ask every time" mode. Agent developers will need to rebuild their first-launch experience and teach their apps to work with reduced permissions.
AI-processed from Habr AI; edited by Hamidun News
A chatbot answers questions and does nothing else. An AI agent works differently: it lives on your computer, opens files, launches other programs, and acts on your behalf. To do all this, such apps often ask for "Full Disk Access." Apple has decided it's time to do something about this, and it has announced changes to Mac's security system that will arrive in macOS 27.3.
Why Apple is acting now
Two high-profile cases triggered the move. A journalist reported that Meta's Muse app gained access to his personal messages without permission. Wired, meanwhile, reported a vulnerability in the ChatGPT app for Mac that could let attackers reach confidential data. In October, Apple said it would revise some of its security settings. Now the company has disclosed the details.
Apple's position is two-sided. The company warns that by granting an app such broad powers, the user exposes everything stored on the Mac to risk. Yet Apple also profits from the trend. AI enthusiasts are actively buying the Mac mini and Mac Studio to run local models. Apple isn't planning to ban agents; it wants to make the risks visible.
What changes in macOS 27.3
First, the settings will be split into two categories: "System & App Protection Override" and "Accessibility." The first explicitly states that the user is turning off built-in protection mechanisms. Previously, the line between these two things was blurry.
Second, granting permissions gets simpler: users just pick the app from a list. Alongside this, recommendations will appear to grant protection-override rights only to programs that genuinely need them. Similar prompts will appear when granting Full Disk Access. Users will get a clearer picture of which data and system functions the app will be able to reach.
Third, control becomes more flexible. An app can be configured to ask for permission on every access. This is inconvenient and gets in the way of full automation, but it allows safer experimentation with local AI systems. Imagine an agent that sorts through your email and documents: instead of one "permanent" checkbox, you confirm each entry into a folder.
Fourth, the system will show a transparency indicator when one app controls another through System & App Protection Override. This closes an unpleasant scenario in which one program quietly steers the others while the user is unaware.
What developers should do
There are no outright bans: agents and other programs with elevated rights can still be installed and run. But Apple asks developers to adapt their onboarding, meaning the first-launch experience. An app should explain why it needs permissions and work correctly under restricted access. If a user chooses "ask every time" or denies full access, the agent must neither crash nor silently lose functionality.
In practice, this means:
- Determine in advance which folders and functions the agent actually needs, and request only those;
- Plan for graceful degradation: what the agent can still do without Full Disk Access;
- Explain the risks and the reasons for the request in plain language during onboarding;
- Test the scenario where permission is requested on every access.
Developers whose product fundamentally requires Full Disk Access, such as for backup or assistive technologies, can request exemptions from the new rules.
Context and limitations
These steps fit into a broader picture. Over the summer, Apple released a major security update for its systems, Safari, and WebKit, closing several dozen vulnerabilities. Some of them were found by researchers using neural-network tools from Anthropic and OpenAI. AI is simultaneously helping to find holes and creating new attack surfaces.
Caveat: this is still an announcement, and the changes will take effect only with the release of macOS 27.3. How strictly Apple will vet exceptions, and how the constant-confirmation mode will behave in real agent workflows, will become clear after release. But the direction is clear: the era of "grant full access and forget about it" for AI agents on Mac is ending.
- Meta is recognized as an extremist organization, and its activities are banned in Russia.
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 2026Apple's macOS 27.3 beta 2 will let you limit AI agents' access to Mac files
- Nov 21, 2026Apple Is Building an AI Agent for Mac While Making It Harder for Outsiders to Get In: 14 Clicks Instead of 4
- Nov 21, 2026Apple Is Recruiting 200,000 Volunteers to Test a Health AI Chatbot
Key Questions Answered
Why Apple is acting now?
Two high-profile cases triggered the move. A journalist reported that Meta's Muse app gained access to his personal messages without permission. Wired, meanwhile, reported a vulnerability in the ChatGPT app for Mac that could let attackers reach confidential data. In October, Apple said it would revise some of its security settings. Now the company has disclosed the details. Apple's position is two-sided.
What changes in macOS 27.3?
First, the settings will be split into two categories: "System & App Protection Override" and "Accessibility." The first explicitly states that the user is turning off built-in protection mechanisms. Previously, the line between these two things was blurry. Second, granting permissions gets simpler: users just pick the app from a list. Alongside this, recommendations will appear to grant protection-override rights only to programs that genuinely need them.
What developers should do?
There are no outright bans: agents and other programs with elevated rights can still be installed and run. But Apple asks developers to adapt their onboarding, meaning the first-launch experience. An app should explain why it needs permissions and work correctly under restricted access. If a user chooses "ask every time" or denies full access, the agent must neither crash nor silently lose functionality.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.