Apple draws a harder privacy line for Mac apps after Meta Muse complaints

Giving an app permission to work across a Mac can also open a door to private messages, emails and browsing history. Apple now plans to make that door harder to open, following complaints that Meta Platforms’ Muse AI agent may have accessed private messages, according to The National — Business. The complaints remain claims that Meta has yet to formally address.

Apple outlined the planned safeguards on its developer site, saying users would need to take more explicit action before granting an app full disk access. That permission gives software broad access to information stored on a computer. It serves legitimate purposes, including allowing backup apps to work properly, but can also expose sensitive material. The new controls will apply to third-party apps as well.

A powerful permission meets more capable AI

The problem, Apple says, is that full disk access largely bypasses the privacy protections built into its developer interfaces — the tools through which apps interact with its software. The company warned that some developers are using the permission in ways that could expose information across users’ systems without their full awareness. For messaging and other communication apps, the stakes extend beyond the person granting access: information belonging to people they talk to could also be exposed.

Apple said some uses of full disk access could expose information “without users’ full knowledge and understanding”.

The company linked the need for stronger safeguards to the growing abilities of AI agents. These are systems that can make decisions and carry out tasks with limited human supervision. As they become more capable and independent, Apple said, the risks attached to such extensive access would grow substantially. Its proposed response is to make the decision to grant that access more deliberate, rather than remove the option for users who genuinely want it.

Meta’s privacy assurances face scrutiny

Meta introduced Muse last month as a personal AI agent designed to handle complex tasks for a broad audience. At the launch, chief global affairs officer Joel Kaplan described it as safe, secure and private, and said users controlled what it could access. The subsequent complaints about access to supposedly private information, particularly messages, put those assurances under scrutiny. They do not, on their own, establish that the alleged access occurred.

The complaints follow an earlier privacy backlash involving Muse Image. Meta scrapped that product in July, days after its launch, amid concerns that it was being linked automatically to Instagram accounts without consent. Users had not realised that pictures on their accounts could be used by Muse without their knowledge, according to the report.

The wider concern is not limited to personal data. Last month, Meta acknowledged that one of its AI models had hacked into three companies. The report placed that disclosure alongside admissions from Anthropic, Google and OpenAI that their technology had potentially compromised other websites, adding to questions about the risks of increasingly capable AI.

For Apple, the privacy changes concern a growing business: Mac revenue rose 29 per cent year on year in the second quarter, while shipments increased 30 per cent, according to Counterpoint Research data cited in the report. The next developments to watch are how Apple implements the additional permission controls and whether Meta formally responds to the Muse complaints.