macOS 27.3 Adds New Security Controls to Stop AI Apps From Accessing All Mac Data

Blog WriterCybersecurity News - Original News Source is cybersecuritynews.com


Apple has introduced new privacy controls in macOS 27.3 beta 2 that make it harder for apps, including AI assistants, to gain broad access to Mac data without clear user approval. Announced on November 16, 2026, the changes add narrower permissions, stronger warnings, and more visibility into app activity.

The update follows Apple’s earlier warning about the risks of unrestricted access, covered in Cybersecurity News’ report on tighter Full Disk Access controls. However, this is not a blanket block on AI tools. Users can still approve broad permissions, and the changes apply beyond AI software.

macOS 27.3 New Security Controls

Apple is dividing Device Control and Data Access into two settings called System & App Protection Override and Accessibility. The company is also introducing a single approval method in which users select the requesting app from a list, making the permission process more consistent across affected workflows.

Instead of pushing users toward unrestricted access, macOS will guide them to approve only the specific apps or folders needed for a task. System & App Protection Override handles app selection, while the updated Full Disk Access flow encourages folder limits. This gives users a narrower alternative to broad approval.

That distinction matters for AI agents that read files and take actions across applications. As earlier reporting on AI browser security risks explains, malicious instructions hidden in outside content can redirect an agent. Limiting its access can reduce the data exposed if it follows those instructions.

According to Apple’s developer announcement, the revised approval screens will explain more clearly which data an app could access or control. Apple warns that granting these permissions puts all Mac data at risk. The goal is informed consent before users approve access that reaches beyond an app’s immediate task.

The supplied screenshots show how direct those warnings can be. A ChatGPT permission screen describes access to messages, email, photos, and browsing activity. A Claude screen offers a choice between overriding system protections and selecting specific apps. These screens describe potential permission scope, not evidence that either app has stolen data.

Users can also configure apps to request approval whenever access is needed. A transparency indicator will show when an app controls another app through System & App Protection Override. MacRumors reports that this indicator appears in the menu bar, helping users notice activity that might otherwise happen out of view.

Existing Permissions Remain Active

One important limit is that previously granted permissions remain in effect. Users will encounter the updated process when approving new requests. Installing the beta therefore does not automatically remove broad access already granted to an AI assistant, automation tool, or another application. Existing approvals still deserve a separate review.

Developers must update setup instructions and ensure their apps work when users approve only selected folders or applications. Apple also plans special entitlements for qualifying backup software and assistive technologies. Known apps currently retain their existing experience without those entitlements, with further details due later.

For security teams, the practical lesson is to match access to the job rather than trust an app’s name alone. These controls can reduce unnecessary exposure, but they do not remove prompt injection or unsafe automation risks. Their value depends on narrow approvals and regular checks of existing permissions.

Give your SOC the alert context to stops incidents quickly. Integrate Threat Intel from 16K SOCs.

The post macOS 27.3 Adds New Security Controls to Stop AI Apps From Accessing All Mac Data appeared first on Cyber Security News.