ChatGPT’s Computer History Logs Your Clicks and Keystrokes on macOS

OpenAI has shipped a new feature called Computer History inside the ChatGPT desktop app for macOS. It records your clicks and keystrokes, builds a personal activity timeline, and makes that timeline available to both ChatGPT and Codex when you ask them something. The goal is to suggest automations, pick up half-finished tasks, and learn how you work. The feature is opt-in, lets you exclude specific apps and websites, and automatically skips incognito or private browser tabs.

What happened

Detail	Fact
Feature name	Computer History
Platform	ChatGPT desktop app, macOS
What it records	Clicks and keystrokes
Models that can access the timeline	ChatGPT and Codex
Opt-in or opt-out	Opt-in
Private browsing	Automatically ignored
Source for private browsing claim	Ari Weinstein, Product and Engineering manager at OpenAI (posted on X)

OpenAI’s Computer History feature sits inside the macOS ChatGPT desktop app and watches what you do on your computer. It converts that activity into a timeline, which both ChatGPT and Codex (OpenAI’s code-focused model) can pull from when you make a request. The practical pitch is that the AI can suggest automations based on patterns it spots, or resume a task you abandoned midway.

Because the feature is opt-in rather than opt-out, it does not activate unless you deliberately switch it on. Once running, you can exclude specific apps or websites from the recording, and you can delete individual timeline entries if you want finer control over what is stored.

What does Computer History actually record?

According to OpenAI, the feature captures clicks and keystrokes and turns them into a structured activity log. That log is what ChatGPT and Codex reference when you ask a follow-up question or request help with something you were working on earlier. Think of it as a memory layer built from your desktop behaviour rather than from conversation history alone.

Ari Weinstein, OpenAI’s Product and Engineering manager, confirmed on X that content viewed in incognito or private browser tabs is automatically excluded. Beyond that, the source does not specify exactly how long entries are retained or whether the data is used to train future models.

Why it matters

This is a meaningful shift in how AI assistants interact with your computer. Most AI tools today work only from what you paste into a chat window. Computer History lets the model reconstruct context from actual desktop activity, which is more powerful but also more invasive.

For business users, the upside is real: an AI that already knows you were halfway through building a report or debugging a script can give faster, more relevant help. The downside is that any keystroke logger, even a well-intentioned one, creates a sensitive data store. If your work involves client credentials, financial data, or anything regulated, you need to know exactly which apps are included before enabling this.

The opt-in design is the right call, and the ability to exclude apps and delete entries gives reasonable controls. But “opt-in” only helps if users read what they are opting into, and that is rarely guaranteed. If you are thinking about how AI tools like this can be integrated responsibly into a business workflow, it is worth reviewing your AI integration approach before rolling out desktop agents to a team.

Our take

OpenAI is building toward a persistent, context-aware assistant that knows your working habits. Computer History is a building block for that, and it is genuinely useful if you do repetitive desktop work. The opt-in default and the app exclusion list show more restraint than we might have expected.

That said, the source leaves out answers to the most important questions: does this data leave your device, is it used for model training, and how long is it kept? Businesses should not enable this feature for employees until those questions are answered clearly by OpenAI. The controls that exist are a good start, but they are not a substitute for a clear data retention policy.

We have seen a similar dynamic in the advertising space, where tools that promise smarter automation often do so by retaining more data than users realise. Our coverage of OpenAI’s recent safety team changes is worth reading alongside this, if you want a fuller picture of the company’s current priorities.

What to do about it

1. Check whether Computer History is enabled in your ChatGPT macOS app settings before assuming it is off.
2. If you enable it, add any app that handles passwords, client data, or financial records to the exclusion list immediately.
3. Review stored timeline entries periodically and delete anything sensitive.
4. Wait for OpenAI to publish a clear data retention and training policy before rolling this out across a team.
5. If you are building automation workflows that incorporate AI context, consider working with a specialist rather than enabling broad desktop tracking by default. Our workflow automation service can scope a safer integration.

The bottom line: opt in with your eyes open, exclude sensitive apps, and hold off on team-wide rollout until OpenAI answers the data retention questions it has not yet addressed publicly.

Source: The Verge · AI