OpenAI swapped ChatGPT's screenshots for keystroke logging, and called it 'friendly'

OpenAI's ChatGPT now records every click, keystroke, and app switch on a Mac that opts in, by user invitation, under a friendlier name.

The new feature, Computer History, shipped in the ChatGPT macOS desktop app on August 13, 2026, via the app's changelog. It replaces Chronicle, an earlier research preview that captured periodic screenshots of the user's screen. Computer History drops the screenshot surface entirely. Instead, it records macOS accessibility interaction events, the system-level stream that includes clicks, typing, keyboard shortcuts, and app switches. The result is a continuous log of what the user is doing, fed into ChatGPT's memory layer and rewritten as plain-text memory files.

The Register's read on the swap is "friendly keylogging." The word names a category security researchers have long flagged: any software that records user inputs at a system level, the same category Microsoft's Recall sat in before its 2024 delay. OpenAI's own documentation does not call it keylogging. It does describe the same surface: events captured locally for up to 48 hours, sent to OpenAI's servers for processing, and deleted afterward. The resulting plain-text memory files stay on the Mac until the user removes them.

The 9to5mac writeup catalogues what the feature does not see: no screenshots, no screen recordings, no microphone input, no system audio, and no macOS Screen Recording permission. Private-mode browsing is excluded. None of that shrinks the data the feature does see. The narrower list of inputs is still a list of every input.

The honest counter to "friendly" lives inside OpenAI's own launch page. The same documentation that introduces Computer History tells users the feature "can increase the risk of prompt injection from malicious content in apps or websites" and recommends pausing collection during conversations with other people without their express consent. OpenAI is admitting that any software it watches can be used to instruct it. A user evaluating the feature is choosing how much of their typing to put inside a system that is already documented to be susceptible to instructions hidden in the data it reads.

Computer History is off by default for individuals. Business and Enterprise workspaces need an administrator to grant access before any member can enable it, and admin access does not auto-opt members in. The feature is available to ChatGPT Pro, Business, and Enterprise customers, requires the existing Memories feature, and does not work with API keys or Amazon Bedrock. It is not available in the European Economic Area, Switzerland, or the United Kingdom. The geographic exclusion is the clearest signal: a feature built around continuous local activity capture is shipping into the same regulatory perimeter where screenshot-based memory has been hardest to defend.

The closest historical reference is Microsoft's Recall, the Windows feature that captured periodic screenshots of the user's screen and was delayed in 2024 after a public backlash over its default-on design. Recall eventually shipped with opt-in mechanics, screenshot exclusion lists, and on-device processing. OpenAI's Computer History lands in a similar posture, with a narrower surface, a more explicit opt-in, and a documentation warning the previous product launch did not have. The critical question is the same: a softer word does not change what gets recorded.

For a reader weighing whether to turn it on, the question is what the feature is allowed to see. Computer History sees every click, every keystroke, and every app switch on a Mac that opts in. The mechanism is the product. The word "history" is the packaging.