Security
Apple splits Mac security settings to counter AI agent risks
New controls in macOS 27.3 give users granular permission options as local AI tools demand broader system access.
Key takeaways
- Apple is splitting Mac security settings into "System & App Protection Override" and "Accessibility" to clarify when users are disabling built-in protections.
- macOS 27.3 introduces per-use permission requests, allowing users to approve AI agent access each time rather than granting permanent authorization.
- The changes follow security incidents involving Meta's Muse app and ChatGPT's Mac application that exposed user data.
- Apple will display transparency indicators when apps control other apps through elevated permissions.
- Developers of backup and assistive technology tools can request exceptions to maintain existing functionality.
Apple is overhauling Mac security settings to address privacy concerns raised by AI agents that require extensive system access to function, according to details first reported by TechCrunch.
The changes, arriving with macOS 27.3, represent Apple's attempt to balance user safety with the growing demand for local AI tools that need deep integration with the operating system to automate tasks across files and applications.
New permission architecture
Apple is splitting the existing "Device Control and Data Access" setting into two distinct options: "System & App Protection Override" and "Accessibility." The redesign aims to make clear that selecting the first option means disabling built-in security protections.
The company is also introducing a unified method for granting permissions through app selection from a list, replacing what had been a more fragmented process. When users enable either System & App Protection Override or Full Disk Access, they'll see new guidance prompting them to grant permissions only to specific apps that require them.
A notable addition is per-use permission requests. Users can now configure apps to ask for access each time it's needed, rather than granting permanent authorization. While this approach may interrupt workflow automation, it allows experimentation with local AI without providing unrestricted access.
Apple will also display a transparency indicator when one app is controlling another through the System & App Protection Override setting, giving users real-time visibility into agent activity.
Why it matters
The security changes reflect a fundamental tension in the AI era: powerful automation tools require broad system access to deliver value, but that same access creates significant privacy exposure. Apple's approach doesn't block AI agents but forces both developers and users to make more deliberate choices about permissions. For enterprises evaluating local AI deployments on Mac hardware, these controls provide a framework for managing risk without abandoning the productivity benefits of agent-based automation.
Context for the changes
Apple announced plans for these updates in October following two high-profile incidents. A journalist reported that Meta's Muse app accessed private messages without consent, and Wired identified a vulnerability in ChatGPT's Mac app that could have exposed sensitive data.
The timing is significant for Apple's hardware business. The company has seen strong sales of Mac mini and Mac Studio systems to early adopters running local AI models, creating a commercial incentive to support these use cases while addressing legitimate security concerns.
In its developer guidance, Apple warned that full disk access "puts all of the data on their Mac at risk" and advised developers to adjust onboarding flows accordingly. The company also suggested that apps should "gracefully handle having access to only a limited set of folders or apps."
Developers creating backup tools or assistive technologies can request exceptions to preserve functionality that predates the macOS 27.3 changes.
Sarah Perez first reported these details for TechCrunch.