On Tuesday, OpenAI also entered the fray with its new AI powered browser, ChatGPT Atlas. The new browser looks a lot like traditional ones such as Google Chrome, but puts ChatGPT at the center of the browsing experience and weaves AI into the user experience. OpenAI is also offering a paid agent mode of Atlas that can conduct searches on its own for users.

For nearly two decades, the basic browsing experience, aside from a few minor improvements, remained largely unchanged. Users typed a url in the navigation bar, or typed a search query in that same space—a feature that Opera first pioneered but which was soon copied by Google—and the browser takes the user to that web address or a search results page, which displays a list of links. Click on a link and the browser takes you to that web page.

Now, tech companies are betting that users want a new kind of experience: a browser that can answer questions, not just provide a list of links, and that can do far more than just navigate a user to a web page—one that can perform tasks for the user on that page, such as booking travel or completing a purchase.

“This is probably the biggest shift since we’ve seen the browser itself become the gateway to the internet. For 30 years, the browser was about navigation. Type, click, explore. Now, with AI, it’s changing the model completely. It’s moving from browsing to delegating,” George Chalhoub, assistant professor at UCL Interaction Centre, told Fortune.

Tech companies, including Perplexity and Opera, have already launched agentic AI browsers that can perform tasks on behalf of users. Perplexity’s Comet combines a web browser with a built-in AI agent that can read pages, summarize information, and even perform multi-step actions, such as booking appointments or sending emails. Similarly, Opera’s Neon introduces features like “Do,” which can carry out actions on a user’s behalf, and “Cards,” which store custom workflows and prompts for repeated use.

“The browser wars are starting and the competition is heating up because browsers today are the operating system of your applications,” Krystian Kolondra, EVP Browsers at Opera, told Fortune. “The browser world is extremely important because it is more aware than the operating system itself about what’s happening on your pages.”

The vision for AI-powered browsing reframes the traditional browser as not just a tool for access, but as the primary interface through which AI agents operate.

“The whole notion of search changed. It’s not to point to the place where you can find your answer or do your thing, but to give you that answer and do that thing,” Himanshu Tyagi, co-founder of the open-source AI company Sentient, said. “We are moving to sort of a dark internet, in the sense that it’s not meant for just humans. It’s meant for bots to consume and process information. Bots do things and give humans the final thing.”

If the third round of browser wars is already underway, the battlefield looks very different from the days when the main axes of competition were things like speed and tab management. This time, it’s about which company can deliver the most seamless AI-powered experience while navigating heightened privacy issues and convincing users to change long-established habits. While major players like Google still dominate, nimble newcomers are testing the limits of what a browser can do.

The third browser wars

The second round of browser wars concluded with Google Chrome’s dominance, mostly due to its speed and integration with the wider Google ecosystem. Most browsers are now also underpinned by Chromium, a free and open-source web browser project, which was primarily developed and maintained by Google. Chromium is essentially a back-end system that determines how a browser goes out to find a particular web address, using an index of web pages Google maintains, and how it renders that web page.

While Google still dominates the search market and has taken steps to integrate AI into the search experience, its market share has been slipping. According to analysts at Third Bridge, in July, Google’s global search market share dropped below 90% for the first time in 10 years.

This could be due to the increased popularity of AI search engines like Perplexity or competition from AI chatbots like ChatGPT, which launched its own Search tool in October last year and announced a full browser on Tuesday. In a survey from brokerage firm Evercore ISI conducted last year, ChatGPT respondents saying ChatGPT was their top search provider increased to 5% from 1% four months prior.

That said, the popularity of Chome as a browser has not noticeably declined. While people may be sending their search queries to ChatGPT or Perplexity, when they use those services on desktop, they are mostly still using a tab in Chrome to do so.

And given the technical complexity and cost of building a browser from scratch, most AI companies are unlikely to develop their own back-end web indexing. Nearly every “AI browser” on the market today, including Perplexity’s Comet and OpenAI’s Atlas, is built on Chromium.

Building a browser entirely from scratch is complex and resource-heavy. To do so, a company would have to recreate everything from how web pages are rendered and memory is managed, to encryption systems, sandboxing, video playback, and constant security patching. Even Microsoft, once Google’s fiercest rival in the browser space, eventually abandoned its own engine and rebuilt Edge on Chromium.

“It’s literally reinventing the wheel,” Chalhoub said. “I don’t see any company building its browser from scratch.”

A unified interface

But why are AI companies so keen to have a browser of their own? Perplexity, for example, raised eyebrows earlier this year when it made an unsolicited cash offer of $34.5 billion for Google Chrome.

“The browser is what we live in during the day on our desktop devices,” Dmitry Shevelenko, Chief Business Officer of Perplexity, told Fortune earlier this year. “It’s just an incredibly powerful canvas, and in terms of being able to create value for users, it gives us a much bigger surface area… it requires us to know more about you and have more context.”

The real prize for these companies isn’t web navigation; it’s control of the gateway to the rest of users’ digital lives, including a lot of other web-based software applications. Most companies are betting that the true value of AI will be unlocked when AI agents have access to a user’s entire ecosystem—emails, calendar, messages, and documents—and can perform tasks across them seamlessly.

“There’s this myth of the ‘everything app,’” Tyagi said. “AI is only magical when it’s everywhere with you in a single, unified interface. If you have to use one app for your glasses, another on your phone, and another on your laptop, that’s not a complete experience. The magic is when there’s one interface that goes with you everywhere and is always engaging with your context.”

Most AI companies are working on autonomous assistants with the aim of getting them moving fluidly between a user’s various applications. But they are pursuing varying approaches to accomplish this.

OpenAI appears to be trying to position ChatGPT as a version of this universal interface, not through a browser, but by integrating third-party apps directly into the chatbot so users can search, shop, plan travel, and manage files without ever leaving the conversation.

But changing user habits is never easy, and the idea of browsing the web is deeply ingrained in users.

“Changing user habits takes time, especially when it comes to something as fundamental as how we explore the web,” Chalhoub said. “For most people, the browser is the oldest and most familiar tool we use online. In terms of physical experience, we trust it so much because it’s stable and predictable.”

However, Chalhoub noted that it’s often small conveniences that drive major behavioral shifts. “If an AI browser can slowly start saving me time, booking travel automatically, or summarizing articles, I think people will adapt much faster than we expect.”

In many ways, AI-enabled browsers, such as Comet, OpenAI’s Atlas, or Gemini in Chrome, are a hybrid, or half-way house, between the idea of chatbots as the universal interface, where a human user has no direct access to the web at all, and the traditional, human-driven web browsing experience. The advantage of an AI browser is that it lets both the human and AI model access the web in exactly the same way—even at the same time, with the agent able to work alongside a person, or work on one process in one tab while the human user works on a different task in another. With the web browser, there’s no need to create an entirely new protocol through which the AI model interacts with the content and data of third-parties, such as Anthropic’s Model Context Protocol (MCP), which is being used to power a lot of “agentic” experiences in chatbots.

The privacy problem

Agentic AI browsers have access to much more user data than traditional search engines, which brings about a host of privacy concerns. By design, these tools see far more of what users do online, and can even infer why a user is acting in certain ways.

“Browsers have always been powerful data collection tools, and when you add AI to the mix, that power multiplies,” Chalhoub said. “An AI-powered browser doesn’t just observe your behavior; it can infer your intentions, habits, and even your mood. Every prompt or summary becomes a data point about you, so the information has to be handled responsibly and not used for advertising or profiling.”

Chalhoub also warned that with AI in the loop, it makes it much harder for users to know where their data goes. “It’s definitely a privacy risk, not because AI is inherently bad, but it has more context and intention in one place. So companies really have to be responsible in this regard,” he said.

It’s already unclear how far a user’s conversations with AI chatbots are confidential. For example, OpenAI CEO Sam Altman recently warned that users currently have no legal protection over their ChatGPT conversations if subpoenaed. Letting an AI agent crawl through emails, texts, and other highly sensitive data could risk exposing deeply personal information if not handled carefully, raising serious questions about how much control users really have over their own data.

Tech companies, for their part, are aware of the new risks around privacy introduced by AI agents. Kolondra said that Opera’s Neon only processes data when users ask it to, like when summarizing a page, and all requests are end-to-end encrypted. “We don’t use that data to train our models,” he added.

An earlier version of this story published Oct. 11, 2025.