ChatGPT Can Now Remember What You Do On Your Computer – But Are We Giving It Too Much Context?

Authored by Mikaelle De Oliveira

There’s a subtle shift happening right now in how we interact with AI. For the past couple of years, prompting on GPT was simple enough: you typed a prompt into a box and it gave you a response based on what you decided to tell it.

Now, that boundary is starting to blur, as tools build memories from our actual behaviour rather than just our prompts.

We already hand over a surprising amount of personal information to chatbots, using them to rewrite awkward texts, settle arguments we probably should have settled ourselves or answer completely random questions at 1 a.m. like whether your cat staring at the same wall for 20 minutes is actually something to worry about (results: unclear).

But, actively choosing to tell a tool is something very different from letting the same tool use information about what you’re doing across selected apps and websites. And that brings us back to the question at hand: are we giving AI too much context?

We’re Still The Ones Giving It Access

To be fair to OpenAI, this isn’t a secret background operation. Computer History is opt-in, meaning users have to switch it on and can change which apps and websites contribute to their history

OpenAI says the feature records interaction events rather than screenshots, screen recordings, microphone input or system audio. Users can also pause the feature and inspect or delete activity from their timeline.

Still, just because something is optional doesn’t mean we should stop questioning whether it’s a good idea. When tracking features arrive behind a simple toggle, it’s easy to accept them without thinking too much about what we’re giving up. That’s why digital privacy still matters, even when the technology is presented as a convenience. In some ways, maybe even more so when it’s presented as a convenience.

It’s Not Just About What We Tell AI Anymore

The bigger change here is behavioural context. Previously, ChatGPT’s understanding was largely based on what you explicitly chose to put into the chat. Now, Computer History can actively use activity from selected apps and websites, including interactions such as clicks, typing and switching between applications, to build a history that ChatGPT can reference later.

That changes the assistant-user relationship. It’s no longer just responding to the questions you type, it can actually gain more insight into how you work. And that feels like a pretty significant step for something that started life as a chatbot.

The Annoying Part Is That It Could Be Really Useful

Of course, there’s a reason people will be tempted to turn this on. If you spend your day moving between Slack, Chrome, Notion and spreadsheets, having an AI that can remember what you were working on could save a lot of time.

You wouldn’t have to keep explaining the same content every time you opened a new conversation. For anyone using AI heavily for work, that sounds genuinely useful. As more tech startups push AI further into workplace productivity, removing that constant need to provide context could become a major selling point.

But that’s also what makes the privacy question harder. The more useful the feature becomes, the easier it is to overlook what you’re giving it access, too.

When Does A Helpful Assistant Become A Spectator?

An AI model remembering a conversation you had last Tuesday is one thing, but an AI building a broader picture from your digital activity feels a little different. It starts to feel less like typing into a notebook and more like having someone sitting beside you, quietly keeping track of what you’re doing throughout the day. Or listening to your conversations and taking notes like a passive aggressive psychologist.

At what point does deep personalisation cross into surveillance? You don’t have to believe the technology is malicious to feel slightly uneasy about where that line sits.

If a platform can understand more about how you spend your working hours, the relationship between you and that tool inevitably changes.

So, How Much Context Is Too Much?

This conversation goes far beyond one OpenAI feature. AI assistants are moving out of isolated chat boxes and becoming increasingly connected to the devices and software we use every day.

As AI developments accelerate, the competition isn’t just about who has the smartest model. It’s also about who can give that model the richest understanding of how people actually work,

Whether you switch Computer History on or leave it off, that direction is becoming harder to ignore. The more AI understands about our routines, the more useful it can become, but we still need to decide how much context we’re comfortable handing over in return. Because convenience is great, but where we draw the line should still be up to us.