ChatGPT Can Now Watch What You Do on Your Mac, If You Let It
OpenAI's new Computer History feature records clicks, typing and app switches on macOS and turns them into memories ChatGPT can use. It is off by default, and the fine print is worth reading.
OpenAI has switched on a feature called Computer History in the ChatGPT desktop app for macOS. With it enabled, the app watches what you do across the apps and websites you allow, turns that stream into written summaries, and lets ChatGPT and Codex refer back to it. You can ask what you were working on before lunch, where that document you opened yesterday lives, or for a summary of your week for a standup.
It is off by default. Pro users can switch it on themselves; in Business and Enterprise workspaces an administrator has to allow it first, and each person still opts in individually. The rollout is global for those plans, with the EU, Switzerland and the UK following in the coming weeks. It needs ChatGPT’s Memories system and does not work through the API.
The technical choice is interesting. This is a rebuilt replacement for Chronicle, an earlier preview that worked from screenshots. Computer History does not take pictures of your screen at all. It reads interaction events that macOS exposes through its accessibility features: clicks, typing, keyboard shortcuts, app switches. According to OpenAI, this captures memories faster and more efficiently, and it means the feature does not need screen recording permission. The raw event stream sits in an isolated container on your Mac, gets processed on OpenAI’s servers into summaries, and is deleted after 48 hours. The summaries themselves are plain Markdown files on your own disk that you can open, edit or delete.
The most consequential part is what the timeline does with all this. When the system spots work you repeat, it can offer to turn that workflow into a reusable skill or a scheduled automation, which Codex then builds from what it recorded. That is the real point of the feature: not a diary, but an agent that learns your routine by watching instead of being told.
Now the caveats, and OpenAI names them itself. The memory files are not encrypted, so other programs running under your Mac account may be able to read them. Because the event stream can include text from websites, the feature widens your exposure to prompt injection, where a malicious page hides instructions that the assistant later follows. And because chat apps can be captured, OpenAI advises turning the feature off during conversations with other people unless they have agreed in advance. Summarising also burns tokens while it runs.
What this means for you: if you are curious rather than committed, this is a good feature to try on a work-only Mac profile with a short list of allowed apps, not on the machine where your banking and private messages live. You can exclude apps, pause from the menu bar, and wipe the last ten minutes, hour, day or everything. If you manage a team, the honest question is not whether it is useful, because it is, but whether your consent practices cover a colleague’s calls being summarised. For everyone else, nothing to do yet, and in Europe not even the option for a few more weeks.
Sources
Source: https://learn.chatgpt.com/docs/customization/computer-history
Qwen3.8-27B Is Out, and It Runs on One Gaming Graphics Card
Alibaba finally shipped the small half of its Qwen3.8 promise. The 27 billion parameter model reached number one on Hacker News because people can actually run it at home.
