Apple Restricts macOS Full Disk Access to Mitigate AI Agent Privacy Risks

Key points

• Apple is updating macOS privacy settings to ensure users explicitly understand the risks of granting Full Disk Access (FDA) to applications.
• The change responds to concerns that AI agents, such as Meta’s Muse, are accessing sensitive data like messages and browsing history without clear user consent.
• Tech columnist Jason Aten reported that Muse accessed his message history despite his belief that he had denied permission, sparking a debate about AI agent security.
• Meta CTO David Singleton argued that Muse requires both FDA and a specific connector setting to access messages, implying users must opt in manually.
• Security expert Patrick Wardle countered that FDA allows any non-root file to be read, contradicting Meta’s claim that access is strictly limited by connector settings.
• Apple’s statement did not name Meta or Muse but acknowledged that FDA is being used in ways that could compromise user privacy and the privacy of their contacts.

Background

This development follows Apple’s recent launch of Siri AI in iOS 27, which faced mixed performance reviews and staggered rollout. It also aligns with Apple’s earlier moves to set usage limits and paid access tiers for Apple Intelligence features, reflecting a broader strategy to manage the risks and costs associated with increasingly autonomous AI tools on its platforms.

How outlets are covering it

Ars Technica emphasizes the contradiction between Meta’s claim that Muse access is opt-in and the technical reality that Full Disk Access allows reading any non-root file, citing security expert Patrick Wardle. Engadget highlights the broader industry trend of AI agents like OpenClaw and Dots requiring FDA, noting that this has contributed to Mac Mini shortages as users seek dedicated machines for AI tasks. MacStories criticizes Apple’s vague announcement, warning that while the intent is to protect users, the changes could inadvertently restrict legitimate non-AI applications that rely on FDA, such as backup tools and utilities like Alfred or PopClip. While all sources agree on the privacy risks, they differ on the potential impact on non-AI software and the clarity of Apple’s implementation plan.

Why it matters

The shift in macOS permissions signals a significant change in how operating systems handle the security implications of autonomous AI agents. By tightening controls on Full Disk Access, Apple aims to prevent unauthorized data exposure, which could influence how other tech companies design their AI integrations. This move may also affect the usability of various Mac applications that rely on FDA, potentially forcing developers to adapt their software to new permission requirements. For users, it highlights the growing need for explicit consent mechanisms when granting AI tools access to sensitive personal data.

What to watch

Apple has not specified a timeline for the rollout of these new controls or the exact nature of the 'additional controls' to be implemented. Users should monitor Apple’s developer blog for updates on how the new permission system will function. Developers of AI agents and other applications that rely on Full Disk Access may need to adjust their software to comply with the new explicit user action requirements. The tech community will likely continue to debate the balance between security and utility, particularly regarding the impact on non-AI applications that depend on FDA for functionality.