Apple has introduced new security controls in its latest macOS beta to give users greater control over how artificial intelligence (AI) applications access files and personal data on their Mac computers.
The update is designed to prevent AI agents from gaining broad access to sensitive information without users fully understanding the permissions they grant. The move follows growing concerns about how increasingly capable AI applications interact with files, messages, browsing history and other private information stored on computers.
Announced on Apple’s developer blog, the new measures add steps before applications can receive Full Disk Access. This permission lets software access a wide range of information across a Mac, making it especially important as AI tools become more autonomous and can perform tasks on users’ behalf.
The controls are intended to make permission requests clearer and more difficult to approve accidentally, particularly when applications ask for access that could expose large amounts of personal information.
Apple strengthens Full Disk Access permissions for AI applications
The new controls come as AI applications increasingly rely on direct access to local files and system information to perform tasks. Applications such as ChatGPT, Claude and Gemini can use different levels of system permissions to help users manage information, search documents and complete other activities.
However, granting Full Disk Access can expose considerably more information than users might expect. Depending on the permissions involved, applications could read, modify, delete, or export personal files, as well as access messages, emails, and browsing information.
Apple said some developers had been using Full Disk Access in ways that could put users’ privacy at risk. The company previously warned that certain implementations could expose sensitive information across an entire system without users clearly understanding the consequences.
Under the revised system, users must take what Apple describes as “very explicit” actions before granting applications this level of access. Rather than relying on a standard permission request, macOS now presents a more detailed warning explaining the potential consequences.
The change is particularly relevant to AI agents, which can perform multiple actions independently after receiving instructions. Unlike conventional applications that typically carry out specific functions, AI agents may interact with different files and services to complete broader tasks. This makes the permissions they receive an important part of protecting user data.
Apple’s approach does not prevent applications from requesting extensive access. Instead, it introduces additional safeguards to help users understand what they are allowing before approving the request.
New warning screen highlights the risks of granting full disk access
The updated Full Disk Access settings include a warning screen that explains in more detail the permissions an application could receive. Screenshots shared by X user @madeby_ldt show that the new interface uses stronger language to communicate the potential privacy risks.
macOS 27.3 now warns you before giving an app full disk access That warning prompt is pretty scary! https://t.co/J2qCKVWJRx
— ldt (@madeby_ldt) November 16, 2026
One warning states, “Your Mac will not be able to protect your data from [your desired AI app],” highlighting the extent of access users are considering granting.
The interface also outlines the specific actions an application could perform. These include the ability to “Read, Change, Delete and Export all your Data”, access hidden and recently deleted photos, and view browsing history while potentially acting as the user.
By presenting these permissions more clearly, Apple aims to make it easier for users to recognise the difference between ordinary application access and permissions that could expose almost everything stored on their computers.
The additional confirmation process is intended to encourage users to consider whether an application genuinely needs such extensive access. It may also help reduce situations where users approve permissions without understanding how much information they are making available.
However, the new controls do not completely prevent AI agents from accessing private information. Users can still grant the requested permissions if they decide the application requires them. The main difference is that macOS now provides a more prominent warning and requires a more deliberate approval process.
The issue gained wider attention in September after Inc. columnist Jason Aten reported that Meta’s Muse Mac application appeared to access his messages despite his belief that he had denied the relevant permission.
Meta responded that Aten must have granted access, maintaining that the feature was entirely opt-in. Nevertheless, the incident raised questions about whether users fully understand the permissions they give AI applications and how those permissions are subsequently used.
Apple addresses growing security concerns surrounding autonomous AI agents
Apple’s latest security measures reflect broader concerns about the risks associated with AI agents that can independently interact with computer systems.
As these tools become more advanced, their ability to perform tasks without constant user supervision could create additional privacy and security challenges. An AI agent with extensive file access could potentially expose sensitive information if it behaves unexpectedly, follows malicious instructions or encounters a security vulnerability.
These risks are especially important for computers marketed for local AI development and processing. Apple’s Mac Studio, for example, is positioned as a machine that can support demanding professional workloads, including running AI models locally.
Although local AI processing can offer privacy and performance advantages, applications and agents running on the device still need appropriate system permissions. If an application receives unrestricted access, locally stored information could remain vulnerable to misuse or accidental exposure.
Apple previously outlined its position on the issue, saying: “As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially. We are committed to ensuring users clearly understand these risks before granting such access so that they can make informed decisions about their own data and privacy.”
The new controls represent a step toward addressing those concerns while allowing users to continue using AI applications that require access to local information.
The feature is currently available in macOS 27.3 beta 2. Apple has not yet announced when the updated Full Disk Access controls will become available to all Mac users through a general software release.
For now, the changes provide an additional layer of protection by making extensive permission requests more visible and requiring users to acknowledge the potential consequences before proceeding.




