OpenAI launches ChatGPT Atlas, a browser with a built-in agent

OpenAI today launched ChatGPT Atlas, a macOS browser that integrates its assistant into web pages and lets users assign it certain tasks. The software is available to free ChatGPT users, although its agent mode is initially limited to Plus, Pro, and Business plans.

Atlas is more than just adding a chatbot to another application. OpenAI wants to occupy the place where users search for information, work and shop online — a position Google Chrome has controlled for years. Versions for Windows, iOS and Android are coming soon, the company said.

ChatGPT stays alongside the open page

Atlas’s most immediate feature is a side panel with ChatGPT. The assistant receives the context of the web page the user is viewing, allowing it to summarize an article, explain a concept, compare information or help draft text without forcing the user to copy and paste text, links and files between windows.

The browser also lets users chat with search results. The approach is similar to Perplexity and Google’s AI Mode: instead of simply displaying a list of links, the system attempts to produce an answer and continue a conversation about it.

This reduces friction, but it also gives the assistant more influence. If ChatGPT selects, summarizes and presents information inside the browser itself, many users may consult original sources less often. For media outlets, retailers and other sites that depend on visits, the question is not just which search engine people use, but how much traffic remains after the AI provides an answer.

Memory based on browsing activity

Atlas can retain information about the pages users visit and the actions they take on them to personalize later responses. That memory could, for example, retrieve a product viewed earlier or resume research without rebuilding the entire context from scratch.

The feature has a clear trade-off: the browser no longer merely records a history of addresses; it can turn some of that activity into context for an AI model. That makes controls over what is remembered and for how long especially important, particularly when Atlas is used for work, to access medical information or to use financial services.

The integration also strengthens OpenAI’s strategic advantage. ChatGPT stops being a tab users visit and starts accompanying them throughout their browsing. The more context it receives, the more useful it may become; at the same time, more digital activity becomes tied to the service.

Agent mode aims to act, not just respond

The most ambitious bet is agent mode, which allows ChatGPT to carry out small tasks inside the browser. Unlike a conventional chatbot, an agent does not merely explain which buttons the user should click: it attempts to navigate a website and complete part of the process on its own.

Web automation remains unreliable. Interfaces change, unexpected pop-ups appear and a simple instruction may require decisions the system does not interpret correctly. In earlier tests cited by TechCrunch, both Perplexity’s Comet and ChatGPT’s agent handled simple requests reasonably well, but struggled with long or cumbersome processes.

There is also a specific security problem: prompt injection. A page can contain text designed to make the agent ignore the original request and follow instructions embedded in the site itself. If the system can browse and take action, a malicious instruction may be more dangerous than an incorrect answer.

That is why it is important to distinguish between assistance and autonomy. Summarizing the open page carries limited risk; allowing an agent to operate connected services, handle private information or confirm actions requires oversight. The launch restricts this feature to paid plans, but the commercial barrier does not eliminate the technical problem.

OpenAI enters Chrome’s territory

Browsers have become a new front in the competition to distribute AI. Perplexity launched Comet and The Browser Company introduced Dia, while Google and Microsoft are adding generative features to Chrome and Edge.

Atlas is therefore entering a market where the idea is already being tested, but with a significant brand advantage: ChatGPT has a user base that could adopt the browser without having to learn a new assistant. Its obstacle is the rival’s scale. Chrome has more than 3 billion users worldwide, according to data cited by TechCrunch, and is integrated into Google’s ecosystem and Android.

Switching browsers also means moving bookmarks, passwords, extensions and habits built up over years. An AI panel can be convenient without being enough to justify the move. The initial lack of versions for Windows and mobile devices further limits the launch’s reach.

The real battle is over who mediates between users and the internet

Google built much of its business by controlling two gateways: search and Chrome. Atlas is trying to combine both in a conversational interface where AI interprets pages, remembers context and can act on them.

Success will not depend solely on the quality of its answers. OpenAI will have to show that its agent completes tasks reliably, that its memory is understandable to users and that the browser can withstand malicious instructions embedded in the web.

Expansion to Windows, iOS and Android will be the next decisive step. Until then, Atlas serves as a test of OpenAI’s strategy: turning ChatGPT into the layer through which people use the internet, rather than merely a service they open within it.