Apple tightens Mac file access as AI agents grow

Apple is tightening the rules on one of the most powerful permissions on its Mac computers, warning that AI agents that can act on a user's behalf create new risks to privacy.

The change affects Full Disk Access, a setting in macOS that lets an app read almost everything stored on a computer, including files, messages, email and web browsing history.

What is changing

Apple says apps will only be able to gain Full Disk Access "with very explicit user action", meaning people will have to make a clear, deliberate choice to give an app that level of access. The company did not set out a precise timetable or say which version of macOS would bring the changes.

Why Apple is acting

"As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially," Apple said.

The company also said: "Some developers are using Full Disk Access in ways that could put users at risk, exposing everything on their systems without users' full knowledge and understanding."

"Addressing this is critical," Apple added. "We are committed to ensuring users clearly understand these risks before granting such access."

What prompted it

The move follows reports of problems with AI tools. A columnist for the American business magazine Inc reported that an AI agent from Meta had accessed his private messages without his permission, and a report by the technology magazine Wired described a potential flaw in ChatGPT's Mac app that could have exposed sensitive data.

What AI agents are

AI agents are programs built on large language models that can carry out tasks on a computer with little supervision, such as reading documents, filling in forms, sending emails or booking appointments. Technology companies including OpenAI, Microsoft, Google, Meta and Anthropic have all launched agents this year.

To work, agents often need access to the same files and accounts as the user. Security experts warn that this creates risks: an agent could be tricked by hidden instructions in a web page or document, a technique known as prompt injection, into revealing private information or taking harmful actions.

Advice for Mac users

• Check which apps have Full Disk Access in System Settings, under Privacy and Security, and remove access from apps that do not need it.
• Be cautious about giving AI tools access to email, messages and files.
• Keep macOS and apps up to date, as updates often fix security problems.
• Only download apps from trusted sources.

The bigger picture

Regulators are paying closer attention to how AI tools handle personal data. In the UK, the Information Commissioner's Office has issued guidance on AI and data protection, and the National Cyber Security Centre has warned that AI is making some kinds of cyber attack easier to carry out.

Apple has long made privacy a central part of how it markets its products, and the change puts it at odds with some developers who want their AI tools to be able to see as much as possible in order to be useful.

Balancing usefulness and safety

The challenge for the industry is that the more an AI agent can see and do, the more helpful it can be, but also the more damage it can cause if something goes wrong. Clear permissions that people understand are seen as one way to strike that balance.