OpenAI is about to shut down its AI browser, which was once thought to be able to disrupt Chrome, but it may have made a mistake from the beginning.

On August 10th, OpenAI will officially shut down its AI browser project, ChatGPT Atlas. With less than a week until its closure, if you are still using Atlas, you need to start looking for alternatives and export your browser's bookmarks, history, and other data to avoid losing access to this data. ChatGPT Atlas, launched in October 2025, didn't even last a year.

About two years ago, "AI browsers" were a very hot niche in the AI ​​industry. Large and small companies, both domestic and overseas, launched products that generated considerable buzz. However, this wave came and went quickly. Now, the popularity of AI browsers has waned considerably, and even OpenAI has decided to shut down ChatGPT.

Atlas has withdrawn from this field. On the surface, this appears to be another casualty of OpenAI's failed attempts in different areas, but considering the overall browser market and AI... If you look into the industry, you'll find that this sector suffers from problems such as fixed user habits, unclear business models, and frequent security issues. In addition, the re-entry of established players in the industry has increased the difficulty of competition in this market. OpenAI wanted to replicate Chrome's success with AI, but... OpenAI spent a year developing ChatGPT Atlas and officially launched the browser last October. The goal is to encourage users to interact more directly with the browser during their daily use, leveraging AI to improve efficiency. The browser itself has built-in ChatGPT, which can read web pages, summarize content, click on behalf of the user, fill out forms, and even plan multi-step browsing tasks. Going back a bit further, the industry already has a large number of products touting themselves as "AI browsers," with varying AI functions and levels of sophistication. It seems that AI is a way to redefine the browser's form. For a time, AI browsers created a buzz in the market, but current research data has poured cold water on the idea. Statistics from research institutions show that by mid-2026, Chrome will still hold approximately 64% of the North American desktop browser market share, Edge approximately 13%, Safari approximately 11%, and Firefox approximately 7%, totaling 95% market share, leaving virtually no room for further growth. Even though many AI browsers have emerged in the past year or two, their combined market share is still far from enough to shake the dominance of Chrome, Safari, and Edge. This means that any product aiming to establish a new entry point by "reinventing the browser" first faces a highly entrenched market: browsers are the type of tool users are least willing to switch. From the user's perspective, Atlas's first problem is migration cost. Browsers handle almost all online activities for daily work and life; bookmarks, history, password vaults, automatic login, plugin ecosystems, and team collaboration tools are all tied to the user's primary browser. For most users, even cramming the most powerful browser into a new one may not be enough to offset the inconvenience of switching their main browser. What we often see is people continuing to use their existing browser and then installing an extension like ChatGPT or Claude, or running an AI client alongside it, rather than migrating entirely to a new browser. In other words, while AI does enhance browsers and bring convenience to users, it's still not enough to justify a user switching browsers altogether. Beyond migration costs, the second layer of problems stems from the security and risk control brought about by AI. The biggest selling point of AI browsers lies in "proxies": allowing models to directly read pages, execute operations, follow links, and retain browsing context for a long time. This also amplifies the security risks of browsers. When Atlas was first launched, security researchers quickly demonstrated several typical problems: for example, attackers could hide malicious prompts in web page content to induce AI assistants to leak sensitive information; they could also use carefully designed links to expose information about websites previously visited by users. Ultimately, while some security issues cannot be directly blamed on OpenAI and are common problems with browsers, for a company that needs to concentrate computing power to build large models, whether the resource investment in a standalone browser to bear such a large security risk is worthwhile needs careful consideration. More importantly, the business logic of standalone AI browsers is unclear. Traditional browsers typically monetize through default search revenue sharing, ad placements, and traffic redirection from their own services. However, OpenAI's real revenue streams lie in API calls, enterprise subscriptions, and the use of high-value models. When Atlas was first launched, we wrote an article hoping that Atlas, as an independent browser, combined with ChatGPT's user base, could break the monopoly of Google Search + Chrome. However, the reality is that Google still firmly holds 91% of the search engine market share, and the recent trend has even shown a slight increase. Unable to capture browser market share or compete in search engine entry points, and unlikely to significantly boost API and enterprise business revenue, the cost-effectiveness is highly questionable considering the computing power and R&D investment. For OpenAI, it would be more reasonable to simply integrate the AI ​​Agent in the browser into a feature of the ChatGPT desktop application. The real winner isn't the AI ​​browser, but the browser transformed by AI. It's worth noting that Atlas's exit doesn't mean "AI in the browser" is unimportant. On the contrary, established players are gradually filling in the gaps in their AI capabilities, building new competitive advantages on top of what they already had. Chrome's AI features are clearly explained on the official website: extracting key summaries from articles, explaining technical terms or concepts, answering questions based on page content, organizing and comparing multiple tabs—a full range of AI functions are available. In addition, Chrome also has an "Auto-Browse" feature, commonly known as the Agent. You simply state your needs, and Gemini will automatically open websites, fill out forms, and complete the process step by step. You can see what it's doing and confirm it at any time. Building upon text, you can also use Gemini Live to chat with it via voice within Chrome. You can use spoken language to discuss ideas and help organize your thoughts, and it will respond with voice. For Google, which already possesses the Gemini model and the Chrome browser, this is a virtually zero-cost enhancement solution. Users don't need to leave their daily Chrome browser; in fact, AI may further increase user engagement.

Although Apple's AI has been criticized for being outdated, Safari has actually embraced AI very actively. Besides common features like webpage summarization found on AI browsers, Safari was also among the first to support MCP. Any MCP-compatible client can connect to Safari through this protocol to access the page's DOM. Structure, network request list, Console output, screenshots, and even interactive operations such as clicking, inputting, and scrolling. To address the lack of extensions in the Safari ecosystem, users can also describe the functions they want to use in Safari using natural language. The system will then call Apple Intelligence to generate corresponding extensions. Users don't need to understand code, APIs, DOM, or extension lists; they only need to describe their needs, and the system will convert those needs into executable steps.

OpenAI initially harbored dreams of rewriting the browser industry with Atlas, but ultimately had to bow to reality. However, OpenAI hasn't completely abandoned browsers. This functionality is currently integrated into its desktop ChatGPT application. If you still want to use ChatGPT in Chrome, you can install an extension, but since it's someone else's domain, the extension's permissions aren't as high, the data it can access isn't as rich, and its functionality is limited. Returning to OpenAI itself, "try - fail - cancel" has gradually become its standard operating procedure. From Sora to Atlas, and then to the standalone Codex application, these products were either halted due to unclear computing power priorities and business paths, or merged into larger product lines. Ultimately, what remained were ChatGPT, enterprise APIs, and workflow platforms—products highly tied to revenue and valuation. Meanwhile, OpenAI also faces commercial lawsuits from Apple, copyright lawsuits from media and publishing institutions, and "security" concerns from regulators and public opinion. Coupled with the pressure of preparing for an IPO and proving long-term profitability to investors, it must make trade-offs in resource allocation, and timely loss mitigation is almost inevitable. With browsers like Chrome, Safari, and Edge already holding a dominant market share, the difficulty for independent AI browsers to leverage user experience advantages to attract mainstream users is akin to reinventing the smartphone. Consequently, the AI ​​browser market has begun to cool down. In the foreseeable future, the browser market structure is likely to remain largely unchanged. We are unlikely to see an AI browser emerge and defeat Chrome; more likely, we'll see "your everyday browser gradually adding more and more AI features." OpenAI's decision to discontinue Atlas marks the end of this product's lifespan and may also signify the end of the AI ​​browser competition.