What is ChatGPT's Computer History feature, and how does it work?
The idea of AI assistants increasingly 'knowing' their users is one of the industry's most visible recent trends. OpenAI's new Computer History feature, added to the ChatGPT desktop app, stands out as one of the most concrete examples of that trend so far.
The feature is currently available only in ChatGPT's macOS desktop app. The core idea is simple: the app turns a user's on-computer actions — which apps they use, which tasks they perform — into a timeline. ChatGPT and its coding assistant Codex can then reference that timeline whenever a user makes a request.
The goal is for the AI to learn how a user works, so it can offer more accurate suggestions, and even help pick up tasks left half done. For example, if a user is midway through editing a spreadsheet and steps away, ChatGPT can recall that context and suggest picking up where they left off.
This kind of feature solves an important engineering problem: until now, AI assistants have effectively worked without memory unless a user re-explained context every time. Computer History fills that gap by passively observing computer activity.
But that kind of tracking naturally raises privacy questions. OpenAI emphasizes that the feature ships off by default, meaning users have to actively opt in — the opposite design, opt-out by default, would likely have drawn far more criticism.
Users can exclude specific apps and websites from Computer History's scope, ensuring sensitive or personal activity isn't tracked. History entries can also be deleted on demand, giving users a degree of retroactive control.
An OpenAI executive said the feature will automatically ignore content in incognito or private browser tabs — meaning at least some sensitive use cases are excluded from its scope by design.
Similar 'context-awareness' features are becoming increasingly common across AI assistants, as different companies try to offer more personalized and proactive help by tracking user activity. But each new feature also means an expansion in the scope of data being collected.
Critics argue that features like this gradually 'normalize' the practice, with users increasingly consenting to broader data sharing without fully realizing it. Supporters counter that opt-in design and granular controls significantly ease those concerns.
Ultimately, Computer History is a concrete example of a broader shift: AI assistants moving from simple question-and-answer tools toward more persistent companions that track a user's digital life in the background. Where that shift ultimately lands will depend heavily on how user trust is managed along the way.
Read next
Why almost every new US power plant this year runs on solar
A close look at every new power plant that came online in the US this year reveals a clear pattern: utility-scale solar leads by a wide margin, followed by battery storage. Fossil fuels trail far behind. Here's what that shift signals about the direction of America's energy future.
What happens when your cloud storage provider disappears with your data?
A US public broadcasting station is at risk of losing access to 50 terabytes of its archive after its cloud storage provider, Iron Mountain, went dark on communication. The case is a stark illustration of how dependent institutions have become on third-party cloud providers — and just how exposed that dependency can leave them.
Anthropic's CEO says the AI backlash is 'fundamentally a crisis of trust'
Anthropic CEO Dario Amodei argues that the growing public backlash against AI stems less from the technology itself and more from a lack of trust in the companies building it. Amodei is also pushing back on criticism that he has been painting an overly pessimistic picture of AI's risks.
What is an AI gateway, and why is Stripe reportedly paying $7B+ for one?
Payments giant Stripe is reportedly set to acquire OpenRouter, a startup that lets developers access dozens of AI models through a single API, for more than $7 billion. Here's what an 'AI gateway' actually does, and why Stripe is making such a large bet on this corner of the AI stack.
What is MCP, and why does it matter that 21,000 servers were exposed
The Model Context Protocol, an open standard that lets AI assistants connect to outside tools and data, has hit a security inflection point after researchers found more than 21,000 unprotected servers exposed on the open internet. Here's what MCP is, why developers adopted it so fast, and why that exposure matters.