ChatGPT can now remember what you did on your computer. Not just what you typed into it. Every click, every keystroke, every app you switched to. OpenAI calls it ChatGPT Computer History, and whether you love it or hate it, you should know exactly what it stores before it runs on your Mac.
What ChatGPT Computer History actually records
The ChatGPT desktop app uses macOS accessibility APIs to capture interaction events: clicks, typed text, keyboard shortcuts, and app switches. It stitches those events into a timeline that ChatGPT and Codex can both read.
What does that look like in practice? In OpenAI’s own demo, an employee asks ChatGPT what the last Google Doc he viewed was. The AI answers with the document’s name. Then he asks if that doc was shared in a group chat with coworkers, and the model digs through his Slack history to check. It got both answers right, which is impressive and unsettling in equal measure.
The point of all this, according to OpenAI, is capability. An AI that knows your recent activity can finish tasks you started, pick up where you left off, and suggest skills or automations based on how you actually work. Think of it as giving the assistant a memory of your workday instead of making you re-explain everything each morning.
For power users, that’s a real productivity unlock. For everyone else, it’s a screen-level recording of your digital life sitting in a file on your Mac. Both things are true at once, which is exactly why Computer History deserves a careful look rather than a quick accept on the permission dialog.
What it ignores, at least by design
OpenAI did build in some guardrails, and they’re worth knowing before you panic.
ChatGPT Computer History is opt-in. It doesn’t turn on by itself, so if you never enabled it, nothing is being captured right now. Activity in incognito mode and private browser tabs is skipped automatically. You can also blacklist specific apps, so your password manager or banking app never gets logged even while the feature runs.
There’s also a memory viewer. OpenAI lets you see the database the AI has built from your activity and delete anything you don’t want in there. That’s more control than some rivals offer, and honestly, it’s the minimum bar this kind of feature should clear.
Sounds manageable, right? Here’s where it gets less comfortable.
The risks nobody mentions in the demo
First, the encryption problem. OpenAI’s own disclaimer warns that ChatGPT Computer History files can contain sensitive information, and that they are not encrypted. Other programs running as your macOS user may be able to read them. So any malware that slips onto your Mac under your user account could potentially rifle through the same timeline ChatGPT uses. That’s a jackpot file for an info-stealer.
Second, the bystander problem. You consented to the tracking. Your colleagues who DM’d you on Slack did not. The demo I mentioned earlier literally shows the AI scanning a person’s Slack message history to answer a question. If a coworker sent something private, it’s now context the model can surface later, and they never agreed to that.
Third, the drift problem. These features start strict and loosen over time. Defaults change, settings get buried, and what was opt-in today has a way of becoming pre-ticked tomorrow. OpenAI’s recent “keystroke collection” headlines show how fast trust erodes when a feature gets described bluntly.
None of this makes Computer History evil. It makes it a trade-off that deserves an informed yes or no.
How to check yours and lock it down
Takes about two minutes. Do this now, whether you plan to use the feature or not.
- Open the ChatGPT app on your Mac and look at the menu bar icon. If Computer History was ever enabled, you’ll see it in the settings under customization, alongside the controls for capture.
- If it’s on and you don’t want it, switch the collection off from the same menu. The menu bar icon is the fastest path; you can pause or stop activity capture without digging through preferences.
- If you’re keeping it on, add a blacklist. Password managers, banking apps, health apps, anything with data you wouldn’t hand a stranger. The feature respects the exclusion list.
- Open the memory view and read what it has already stored. Delete anything that made you wince. Don’t just scroll, actually remove entries.
- Check it again after updates. Feature behavior shifts between versions, and a two-minute audit once a month keeps you in control.
This same checklist works as a mini privacy routine for any AI tool with activity capture, and more of them are adding it every quarter. If you want a second layer of protection for chats themselves, we already have a walkthrough of locking your ChatGPT conversations with a PIN.
Recall, but different (and that difference matters)
If this sounds familiar, it’s because Microsoft tried something like it first. Windows Recall takes continuous screenshots of your desktop and uses AI to make everything searchable. The first version was a security disaster, storing sensitive screenshots in an unencrypted folder that researchers could read within hours of the announcement.
OpenAI’s version is not a screenshot machine. It records events and interactions instead, clicks and keystrokes and app switches, which is technically less invasive than photographing your entire screen. Here’s the comparison that matters:
| Windows Recall | ChatGPT Computer History | |
|---|---|---|
| What it captures | Continuous screenshots | Clicks, keystrokes, shortcuts, app switches |
| Where analysis happens | On-device AI scanning | Timeline used by ChatGPT and Codex |
| Private browsing skipped | Yes, in current version | Yes, automatically |
| Per-app exclusions | Yes, can filter websites and apps | Yes, manual app blacklist |
| Storage encryption | Encrypted after early controversy | Not encrypted, per OpenAI’s own docs |
| Default state | Opt-in | Opt-in |
That unencrypted storage row is the one to remember. Recall got destroyed publicly for less, then patched it. OpenAI is shipping with the disclaimer up front, which is honest, but honesty isn’t encryption.
There’s a bigger backdrop here too. AI assistants keep asking for more access to your stuff, and the failures pile up when companies move fast. The Notion AI data breach showed how a third-party integration can leak private transcripts, and it’s the same pattern: convenience first, security later. You are the one who has to set the boundaries.
The bottom line for regular users
You don’t need to be a privacy zealot to make a good call here. The decision rule is simple: does the productivity gain beat the blast radius if that file leaks? ChatGPT Computer History is powerful, but power always has a price tag attached.
If you handle client secrets, financial data, legal documents, or anyone else’s private information on your Mac, the answer is probably no for now. Turn it off, blacklist the sensitive stuff if you experiment later, and revisit when OpenAI encrypts the storage. The feature will still be there in six months, and it’ll likely be safer.
If your work is low-sensitivity and you genuinely benefit from an assistant that remembers your workday, opt in deliberately. Blacklist the obvious apps, audit the memory view weekly for the first month, and keep the capture paused on days when you’re handling anything you wouldn’t want in a timeline. You can read OpenAI’s official Computer History docs for the exact settings, and it’s worth doing before you flip the switch. For the reporting behind this piece, see Futurism’s original coverage.
The takeaway: opt-in means the choice is yours, and it’s a real choice with real trade-offs on both sides. Check your settings today, decide on purpose, and don’t let a dialog box make it for you.