OpenAI's Computer History feature turns ChatGPT into a full-time observer

Picture this: you're deep into a spreadsheet, then you hop over to Slack, then you check a flight booking site, then you go back to finish an email. A week later, you open ChatGPT and ask it to "pull up that flight I was looking at" — and it just knows. No re-explaining, no digging through tabs. That's the pitch behind OpenAI's newest and most debated feature yet.

On August 13, 2026, OpenAI quietly rolled out something called Computer History inside the ChatGPT desktop app for Mac. It quietly builds a searchable timeline of what you click, type, and switch between, then feeds that back to ChatGPT for more context later. That's a massive shift from how AI memory has worked until now — instead of only remembering what you explicitly tell it, ChatGPT can now build a picture of your day just by watching you work.

Naturally, the internet lost its mind a little. Comparisons to Microsoft's widely criticized Recall feature started almost immediately, and for good reason — this is the closest any major AI assistant has come to full-on "watch everything you do" tracking. So let's break down exactly what it does, what it doesn't do, and how to control it before you ever consider flipping the switch.

What is Computer History and how does it actually work?

Computer History lives inside the ChatGPT desktop app for macOS, and you manage it through the app's Settings menu under Integrations. According to OpenAI's own Computer History documentation, the feature builds on top of ChatGPT's existing Memories system, converting your day-to-day activity into text-based memory files that ChatGPT and Codex can reference later.

Here's the technical part that matters: Chronicle relied on screenshots. Computer History instead records interaction events exposed through macOS accessibility features, including clicks, typing, keyboard shortcuts, and switching between apps. That's a meaningful design change from OpenAI's earlier "Chronicle" research preview, which took periodic screenshots and ran OCR on them.

Instead of pictures of your screen, Computer History captures a stream of interaction events — essentially a structured log of your behavior — and turns that into periodically turns that activity into text summaries organized in a timeline by day and time that can be used to resume previous tasks, locate recently accessed files, summarize work, or identify repeatable activities that could become skills.

Importantly, OpenAI is explicit about what it does not capture. It records interaction events—not screenshots, screen recordings, microphone input, or system audio—and private browsing is not included.

Where does all this data actually go?

This is where things get genuinely interesting from a privacy standpoint, and it's the part most coverage glosses over. The Sam Altman-led company states that interaction data collected via Computer History is not used to train OpenAI models. The generated memory files are sent back and stored as plain text directly on the user's Mac for local processing.

But the data doesn't stay purely local the whole time. Interaction events are intended to be stored locally for up to 48 hours before ChatGPT and Codex delete them, according to The Register. The events are sent to OpenAI servers for processing into memories. OpenAI says it does not retain the event files after processing unless required by law and does not use them for training.

Once the memories are generated, though, they sit on your machine — and not in a locked box. Concerns have been raised because the memory files stored on the user's Mac are unencrypted, meaning other local programmes with access to the file system could potentially read them. If you share a Mac, use shared login credentials, or have other software with file system access, that's worth sitting with for a second.

There's also a subtler risk that gets buried in the fine print. Computer History increases the risk of prompt injection from content in apps and websites. For example, if you visit a website containing malicious instructions, ChatGPT or Codex might follow those instructions. In plain terms: if Computer History reads a webpage that contains hidden text designed to manipulate an AI, ChatGPT could theoretically act on it without you ever seeing that instruction.

Why does OpenAI want you to pause it during calls?

Buried in the opt-in agreement is a line that raised more than a few eyebrows. OpenAI's own guidance states plainly: "Turn it off during communications with other people unless you have their prior express consent," the company advises, perhaps in acknowledgement of legal risk. "Consider pausing it or excluding apps that contain sensitive health, financial, or personal information."

Think about what that actually means in practice. If you're on a video call, in a group chat, or texting a friend, Computer History could be logging that interaction — which means you're technically supposed to get consent from everyone on the other end before letting it run. Realistically, almost nobody is going to pause an AI feature every time they open Slack or join a Zoom call, which is exactly why this clause exists — to cover OpenAI legally rather than to solve the actual problem.

One hands-on tester summed up the tension well, noting "I doubt that very many people will heed the warning and comply even though doing so is just a click of the menu bar away."

Who actually gets access to Computer History right now?

This isn't rolling out to every ChatGPT user. As the feature is strictly opt-in, the feature remains off by default until a user manually enables it. The Computer History feature is currently rolling out to ChatGPT Pro, Business, and Enterprise users. For Business and Enterprise accounts, administrators must approve the feature before individual team members can activate it.

If you're on a company Mac, there's an important wrinkle worth knowing about. Personal ChatGPT Pro accounts can turn Computer History on themselves, even on a company-owned Mac, with zero visibility for the employer's IT team; Business and Enterprise seats need an admin to turn it on first through Workspace Settings. That means an employee could technically enable this on a work laptop without their IT department ever knowing — a detail that's likely to make a lot of security teams nervous.

Geography matters too. The Computer History feature is currently being rolled out globally. However, users in the European Economic Area, the UK, and Switzerland are excluded from it, receiving the feature in the coming weeks. That delay almost certainly comes down to stricter data protection regulations like GDPR, which require more scrutiny before shipping something this invasive.

How do you turn on (or control) Computer History?

If you want to try it, here's the actual path based on OpenAI's own instructions:

1. Update to the latest ChatGPT desktop app for Mac.
2. Go to Settings, then Integrations. To use Computer History, opt in under Settings → Integrations in the ChatGPT desktop app on Mac.
3. Choose which apps and websites you want included. Choose which apps and websites contribute, pause collection, and review or delete your history at any time.
4. Review your timeline any time from Settings > Computer History > History. Open Settings > Computer history > History to inspect what Computer History has summarized. You can reveal a summary's local memory file in Finder, delete an individual timeline item, or clear the last 10 minutes, last hour, last day, or all history.

If you decide it's not for you, disabling is just as simple — go back into the same settings menu and toggle it off. No opt-in means no data collection.

Before you even opt in, it's worth building your exclusion list first. Password managers, banking apps, health portals — anything sensitive should be off-limits from day one rather than something you clean up after the fact. Choose only the sources you need. Under Computer History permissions, use Include only these apps and Include only these websites for the pilot. Exclusions are useful later, but a short allowlist makes the initial data boundary visible.

Is this really just OpenAI copying Microsoft's Recall?

Not exactly, and the differences matter. As Computer History logs granular interaction events, industry experts have drawn comparisons to Microsoft Windows' controversial Recall feature. But Recall took full screenshots of your screen every few seconds — Computer History captures structured interaction events instead, which is arguably less invasive in one sense (no visual record of everything on your screen) but potentially more precise in another (it directly logs your keystrokes and clicks rather than inferring from images).

The bigger context here is competitive pressure. AI companies are racing to make their assistants feel less like chatbots and more like an actual second brain that already knows what you're working on. Persistent, ambient memory is the next frontier after chat history and file uploads — and whoever nails it first gets a serious retention advantage. That's likely no coincidence given how aggressively competitors have been pushing memory and personalization features of their own recently.

What should you actually do about it?

If you're intrigued by the productivity upside, there's a real case for cautious experimentation — start with a tight allowlist of only a couple of low-stakes apps, check your timeline daily for the first week, and get comfortable with the pause and delete controls before trusting it with anything sensitive. If you're someone who values privacy above convenience, sitting this one out entirely and waiting to see how the first few months play out is a perfectly reasonable call too.

Either way, the fact that OpenAI shipped this before nailing down encryption on the local memory files tells you something about the pace this industry is moving at. Convenience is winning the argument for now — but the tradeoffs aren't going away just because you can't see them.