OpenAI has rolled out a new feature called “Computer History” for the ChatGPT Mac app, turning clicks, keystrokes, and app switches into a searchable timeline. It replaces Chronicle, the screenshot-based experiment introduced back in April, and is meant to give ChatGPT and Codex a better memory of your workday. Turn it on and you get real convenience, but you also sign up for a data-handling practice that OpenAI itself openly flags as a weak point.
Key takeaways
- Computer History replaces the screenshot-based Chronicle feature and instead captures interaction events through macOS’s accessibility system.
- It logs clicks, keystrokes, keyboard shortcuts, and app switches, but no screenshots, microphone input, or system audio.
- The locally stored event and memory files are, by OpenAI’s own documentation, unencrypted; other programs running under the same Mac user account could in theory access them.
- Whether the data gets used for training depends on a user’s individual ChatGPT data controls, it isn’t automatic.
- Available initially to Pro, Business, and Enterprise users; excluded are users in the European Economic Area, Switzerland, and the United Kingdom.
From screenshots to an event stream
Chronicle, launched as an experiment in April 2026, worked by taking periodic screenshots to give ChatGPT context on whatever the user was doing, technically simple, but a privacy headache, since any given screenshot could show literally anything on the screen. Computer History takes a different route: through macOS’s accessibility interface, the app reads structured interaction events from approved apps and websites, periodically turns them into text summaries, and builds local memory files sorted by day and time. If you got pulled away from a task, you can later just ask ChatGPT what you were last working on instead of re-explaining the context by hand. OpenAI stresses that it captures neither screen content nor audio, only structured event data.
The unencrypted weak spot
This is exactly where things get technically uncomfortable. OpenAI’s own documentation states plainly that the event and memory files are “not encrypted” and that other programs running under the same macOS user account could in theory access them. Temporary event files stick around for up to 48 hours, and the memory files generated from them persist until manually deleted. They do sit inside an isolated app group rather than out in the open file system, but for a feature that is, at its core, a searchable log of your entire workday, skipping encryption is a deliberate choice with real risk attached. OpenAI also warns of an elevated prompt-injection risk on top of that: malicious content inside a monitored app or website could manipulate the logged events. Whether the collected data then also gets used to improve OpenAI’s models isn’t a flat yes-or-no either, it depends, per OpenAI, on a user’s individual data controls inside their ChatGPT account, so anyone who hasn’t checked those settings should probably go look at what’s actually turned on by default.
To be fair, OpenAI does offer fine-grained control over what gets logged in the first place. Users can individually include or explicitly exclude specific apps and websites, pause and resume recording at any time, and delete the entire history collected so far. On Business and Enterprise accounts, administrators additionally have to enable the feature for their organization before individual employees can even turn it on themselves, a two-step opt-in that deliberately slows the rollout down. Nobody ends up with Computer History running automatically: do nothing, and nothing gets logged.
Why Germany is sitting this one out for now
The geographic restriction stands out: users in the European Economic Area, Switzerland, and the UK aren’t even offered Computer History in the current rollout. That follows an increasingly familiar pattern among US AI companies: launch data-hungry new features first wherever regulatory requirements are lowest, and bring the European market in later, once privacy impact assessments and GDPR-related adjustments are done. For German readers, that’s both an advantage and a preview: anyone who wants to try the feature now would need an international account, but will very likely face the same trade-off before too long, the same one US users are weighing right now.
Bottom line: convenience traded for control
Computer History is a clean example of where AI assistants are heading: away from a plain chat window and toward systems that keep track of your entire digital day, so you have to re-explain context less often. The convenience gain is real, especially for people who lean on ChatGPT and Codex throughout their actual workflows. The price is a detailed, locally stored behavioral log that, by OpenAI’s own admission, isn’t adequately locked down against other software running on the same machine. Anyone turning this on should go in aware that they’re trading convenience directly for control over their own data, and it’s worth actually checking the data-control defaults before flipping the switch.