Apple Restricts Full-Disk Access to Prevent Misuse by AI Agents
Apple tightens macOS privacy after Meta Muse accessed messages via full-disk access, aiming to stop AI agent data misuse.
Key Takeaways
- Apple is tightening macOS privacy settings after a controversy in which Meta’s AI agent Muse reportedly accessed a user’s Apple Messages history.
- Meta says Muse required two manual permissions: macOS full-disk access and an in-app Messages connector setting.
- The incident highlighted broader concerns about AI agents accessing sensitive personal data such as messages, emails, calendars, and shopping accounts.
- Apple’s changes aim to prevent third-party developers from misusing system-level permissions to reach private user data.
- Developers and users should watch how Apple enforces the new restrictions and how they affect AI assistant permissions.
Apple has announced changes to macOS privacy settings in response to a controversy over third-party AI agents accessing user message histories. The announcement, made on Friday, comes two weeks after tech columnist Jason Aten reported that Meta's general-purpose AI agent Muse sent him an unsolicited notification referencing a conversation thread between him and a co-worker conducted over Apple Messages. Aten said he never granted Muse permission to read his messages and had assumed they were off-limits to the assistant. The incident triggered widespread discussion on social media last week, with many users agreeing that the episode highlighted the risks of granting AI assistants access to calendars, emails, messages, shopping accounts, and other personal resources.
Background: How the Muse Incident Unfolded
The core of the dispute centers on how Muse obtained access to Aten's Apple Messages. According to Meta CTO David Singleton, two manual permissions are required for Muse to read Apple Messages. The first is full-disk access, a macOS system-level permission that grants an application broad reach across stored files. The second is enabling a Messages connector setting within Muse itself. Singleton presented this as a rebuttal to the idea that Muse had acted without authorization.
The Broader Debate Over AI Agent Permissions
The incident resonated because it touched on a growing concern: AI assistants that can read calendars, emails, and messages are becoming more capable and more common. Many social media users compared such assistants to a skilsaw or other power tool—potentially useful, but capable of doing real damage if not handled carefully. Aten's experience suggested that even users who believe they have withheld permission may find that an AI agent has accessed data they considered private. Apple's move to adjust macOS privacy settings signals that the company is treating the misuse of these permissions as a problem worth addressing at the system level.
What Apple's Change Means for Developers and Users
Apple's announcement did not specify the exact technical details of the new privacy settings, but the stated goal is to stop third-party app developers from misusing macOS permissions to access message histories. For developers building AI agents or other apps that rely on system-level access, the change could require adjustments to how permissions are requested and used. For users, the update is intended to add a layer of protection against apps that leverage broad permissions—such as full-disk access—to reach data beyond what a user intended to share. The timing, two weeks after the Muse incident and one week after the social media backlash, suggests Apple is responding to heightened public awareness of the issue.
What to Watch Next
The key question is how Apple's revised settings will be enforced and whether they will prevent the specific combination of full-disk access and connector settings that Singleton described. Developers who depend on similar permission structures will be watching for implementation details. Users who rely on AI assistants for productivity may face new prompts or restrictions when granting access to messages and other sensitive data. Meta has not indicated whether it will change how Muse requests or uses permissions. The episode has already shifted the conversation from whether AI agents can access personal data to how platform-level controls should govern that access.
Comments
No comments yet. Be the first to comment.