Apple is tightening one of macOS's most sensitive permission settings as the company moves to curb the misuse of AI agents that can scrape personal data from a user's computer. The change affects full-disk access, a powerful privilege that allows software to read broad swaths of files, messages, and other content stored locally on a Mac. In practice, the adjustment is designed to make it harder for automated tools and agentic AI systems to quietly collect information that users may not realize they are exposing.
The move lands at a moment when the technology industry is racing to embed AI assistants deeper into operating systems, browsers, and productivity tools. Those systems often need access to calendars, emails, documents, and messages to be useful. But the same access that makes them capable also makes them risky. Apple's latest posture suggests it is drawing a firmer line around the most sensitive data categories, even as competitors push for broader permissions in the name of functionality.
Privacy Guardrails
Apple has long positioned privacy as a core product differentiator, but the rise of AI agents has created a new test for that promise. Traditional apps usually request access for a narrow purpose, and users can often understand the trade-off. AI agents, by contrast, may operate continuously, infer intent, and traverse multiple data sources at once. That makes full-disk access especially fraught, because once granted, it can become a catch-all pathway to highly personal information.
The company's revised approach appears aimed at reducing the chance that a single permission prompt becomes a blanket authorization for machine-driven surveillance of a user's digital life. For Apple, the issue is not only malicious software. It is also the possibility that legitimate AI products could overreach, collecting more than users intended or more than the task requires. In a market where AI features are increasingly marketed as seamless and autonomous, Apple is signaling that convenience cannot come at the expense of granular control.
Meta Pushes Back
The policy shift also sharpens a dispute with Meta, which has argued that Apple's existing framework is not sufficient for its Muse reading messages feature. Meta's position reflects a broader industry frustration: developers want enough access to make AI assistants genuinely useful, while platform owners want to prevent those same tools from becoming privacy liabilities.
Apple, however, appears unconvinced that broader access is justified. The company's stance suggests that it believes the burden should remain on developers to design systems that work within tighter constraints rather than asking users to surrender more of their data. That distinction matters. If Apple prevails, AI products on its platforms may need to rely more heavily on on-device processing, selective permissions, and narrower data pathways. If Meta's view gains traction, the industry could see pressure for looser access rules under the argument that advanced AI requires deeper inspection of user content.
The disagreement is emblematic of a larger strategic divide in Big Tech. One camp sees AI agents as software that should be empowered to act broadly on a user's behalf. The other sees them as tools that must be boxed in by default, especially when they can read messages, files, and other intimate records. Apple's latest move places it firmly in the second camp.
Wider Industry Stakes
The implications extend beyond Apple and Meta. Cloud providers, chipmakers, and software developers are all betting on a future in which AI systems can navigate personal and enterprise data with minimal friction. But the more autonomous those systems become, the more they resemble privileged operators rather than ordinary apps. That raises questions for regulators, security teams, and consumers about how much access is appropriate, how consent should be presented, and what safeguards should exist when software can act at scale.
For Apple, the change also reinforces a familiar business logic: tighter permissions can preserve trust in the platform, even if they complicate development. For rivals, the challenge is to prove that AI agents can deliver meaningful utility without becoming invasive. The outcome of that contest will shape not only the next generation of consumer AI features, but also the standards that govern how much of a user's digital life software is allowed to see.
In the near term, Apple's move is likely to force developers to revisit how they request access and how they justify it. It may also slow the rollout of some AI features that depend on broad local data visibility. But the broader message is unmistakable: as AI agents become more capable, the permissions that govern them are becoming a frontline privacy issue, not a back-end technical detail.