Apple tightens macOS permissions to curb AI agent data risks

Key points

• Apple announced new safeguards for the 'Full Disk Access' setting in macOS to address privacy risks posed by AI agents.
• The company stated that some developers are using this permission to access files, emails, messages, and browsing history without users' full understanding.
• Future updates will require 'very explicit user action' to grant apps this level of access, though specific implementation details and timelines were not disclosed.
• The move follows reports that Meta's Muse agent accessed private messages on a user's Mac despite the user believing they had denied permission.
• Apple emphasized that as AI agents become more capable, the risks associated with broad system access will grow substantially.

Background

This development follows the recent launch of iOS 27 and the introduction of more capable Siri AI features, which have increased the integration of AI assistants across Apple devices. Earlier this month, Apple also detailed usage limits and paid access tiers for Apple Intelligence features, signaling a broader strategy to manage the resource and privacy implications of advanced AI tools. The current push for stricter disk access controls aligns with Apple's recent efforts to position itself as a privacy-focused leader in the competitive AI agent market, where companies like OpenAI and Meta are deploying persistent digital proxies that require deep system integration.

How outlets are covering it

While all outlets report on Apple's announcement, they emphasize different aspects of the issue. Engadget highlights the potential for developers to mislead users about privacy tradeoffs and notes that the demand for dedicated AI machines has contributed to Mac Mini shortages. MacRumors focuses on the technical details of the announcement, noting that Apple did not specify when the changes would roll out or exactly how the new controls would function. The Verge frames the issue as a response to specific privacy breaches, citing a report by Jason Aten that Meta's Muse agent accessed his messages despite his belief that he had not granted permission. Business Insider notes that Apple did not specify if the changes would apply exclusively to AI agents, but rather to any app seeking full disk access as agents become more autonomous. There is a slight disagreement on the scope of the issue: while Apple's statement implies a broad risk, some outlets like The Verge and Business Insider focus heavily on the specific controversy involving Meta's Muse, whereas Engadget points to a wider trend among various AI developers.

Why it matters

This change is significant because it directly impacts the privacy and security of millions of Mac users who are increasingly adopting AI agents that require deep system access. By requiring 'very explicit user action' for full disk access, Apple aims to prevent unauthorized data exposure and ensure users are fully aware of the risks involved. This move could set a new standard for how operating systems handle permissions for autonomous AI tools, potentially influencing how other tech companies design their privacy controls. It also highlights the growing tension between the capabilities of AI agents and the need to protect user data, a key concern as these tools become more integrated into daily life.

What to watch

Apple has not disclosed a specific timeline for when the new 'Full Disk Access' controls will be implemented or the exact nature of the changes. Users should expect updates in future macOS releases, likely tied to the upcoming macOS Golden Gate update or subsequent patches. Developers of AI agents and other software that rely on full disk access will need to adapt their applications to comply with the new, more explicit user consent requirements. The tech industry will likely watch closely to see how these changes affect the adoption and functionality of AI agents, and whether other operating systems follow suit in tightening permissions for autonomous software.