Apple has announced changes to macOS security controls aimed at AI agents that ask for full disk access. The update appears in the latest beta of the system and affects apps such as ChatGPT, Claude and Gemini, which typically request this level of permission to act on the user's behalf.
The change follows a warning the company itself issued last month about the risks of granting this kind of access. According to the source, Apple now requires users to take "very explicit" steps before approving this extraordinary permission. The company has also split device control and data access into two separate settings, to make the difference between the two clearer.
In practice, Apple says users must understand, before granting access, that it exposes all the data stored on the Mac. The source stresses that the warning is aimed at app developers, who will now have to clearly explain why they need this permission.
For those who manage teams using Macs, the immediate takeaway is to review which AI apps already have full disk access and whether that grant is still justified. Since the feature is in beta, the final behavior may change before the official release. The paper considers the move the right one: broad permission for an agent that carries out tasks without constant supervision demands an explicit justification and should not be granted by default.
