ChatGPT logs your Mac activity, and the files are not encrypted

OpenAI switched on a feature in the ChatGPT app for macOS on August 13 that keeps a record of what you are working on. It is called Computer History, it collects your activity across apps and websites, and it turns that into a searchable timeline plus lasting memories that ChatGPT can draw on later. The feature is off by default and limited to Pro, Business and Enterprise accounts.

What it records, and what it leaves alone

Computer History captures interaction events: clicks, typing, keyboard shortcuts, app switches and the context macOS exposes through its accessibility system. It does not take screenshots or screen recordings, and it never touches the microphone or system audio. Private browsing is excluded, and the feature does not ask for Screen Recording permission at all. That is the break with its predecessor, because Chronicle, the research preview it replaces, worked from screenshots. Every so often the events are condensed into text summaries, and when ChatGPT spots a task you repeat, it offers to turn it into a reusable skill or an automation.

Local, up to a point

The raw events sit on your Mac inside the app's own container and are deleted after 48 hours. They still leave the machine before that. OpenAI processes the temporary event files on its servers to generate the memories, then stores the finished memories back on your Mac. Those files are not retained afterwards and are not used for training, the company says.

That changes once a memory is actually used. Contents pulled into a chat can feed model improvement, depending on your ChatGPT data controls. Computer History also requires Memories, the feature we put side by side with Gemini and Claude.

OpenAI flags two risks itself

The first one sits in its own documentation, in plain words: the files can hold sensitive information, they are not encrypted by Computer History, and other programs running as your macOS user may be able to read them. OpenAI's advice is to secure your Mac account and exclude anything sensitive up front.

The second is prompt injection, meaning instructions hidden inside content the assistant reads. Visit a page carrying those instructions and ChatGPT may follow them, and Computer History raises that risk, OpenAI says. We saw what that looks like in practice when five AI browsers were hijacked by a single email. There is also a line many will skim past: the feature should stay off during conversations with other people unless you have their express consent.

Turning it on and reining it in

In a Business or Enterprise workspace an admin grants access first, and even then nobody is enrolled automatically. Each person opts in individually. In the Mac app the path runs through Settings to Integrations and Computer history. From there you decide which apps and websites may contribute, either as an exclusion list or as a narrow allow list. The menu bar icon pauses collection. Deleting works in steps, from the last ten minutes up to everything, and it removes the memories built from those events as well. That cannot be undone. If you are going through ChatGPT's settings anyway, there are more switches worth knowing.

Not everywhere yet

OpenAI's release notes state that Computer History is not currently available in the European Economic Area, the United Kingdom or Switzerland. No date is attached. Several reports mention a follow-up launch within weeks, but that cannot be sourced back to OpenAI. One more cost to factor in: the summaries consume tokens, so they count against your allowance.