Why Every AI Company Is Racing to Become Your Browser
OpenAI launched the AI browser that was supposed to dethrone Chrome, and then killed it. Atlas stopped working on 9 August 2026, 292 days after it launched, and its features moved into ChatGPT and Codex.
If AI browsers are the future, why did the most talked about one disappear? Because the standalone browser was never the point. The point is the browser session, the place where you are already logged into your email, bank, calendar and shopping accounts. Whoever controls that session controls what an AI agent can actually do for you. Atlas lost as a product, but the race for that session is accelerating.
This article explains who is racing, what each company is really after, and what it means for your accounts and your data.
The Browser That Lasted 292 Days
Atlas looked like the boldest bet in the category. It launched on 21 October 2025 as a Chromium based desktop browser with ChatGPT in a sidebar, for macOS only at first.
The retreat came quickly. OpenAI said it would move the browser into the new ChatGPT desktop app, so browsing becomes part of existing AI workflows. The company was consolidating, not quitting. A separate browser was one more app to switch to, and OpenAI wanted one app.
There was a practical cost for users. Atlas bookmarks would not transfer automatically, and the discontinued browser would stop receiving security updates.
The lesson: the AI company did not give up on browsing. It decided the browser should live inside the assistant, not the other way around.
Why the Browser Is the Prize
An AI agent that can only read public web pages is a research tool. An agent that can click, type and pay inside your logged in session is a worker. That is the difference everyone is chasing.
Three things make the browser valuable:
- Your logged in state. Cookies, saved passwords and open accounts are the access layer for most of your digital life.
- Distribution. Chrome held roughly 70% of the browser market when the AI browser rush began, with every rival below 15%. Whoever sits in that window gets the first look at every task.
- Discovery. When you open a new tab in an AI browser, you see an agent instead of a search bar, so the question of whether to Google or use AI fades.
The barrier to entry is also low. Nearly every AI browser on the market, including Comet and Atlas, is built on Chromium. Anyone can ship a browser. Winning the session is the hard part.
Who Is Actually Racing (and How)
Each company is approaching the same prize from a different side.
Google is defending its territory. On 28 January 2026, Google added an agentic mode to Chrome called auto browse. It handles multistep chores such as forms, scheduling and shopping, and started for AI Pro and Ultra subscribers in the US. The reach is growing: on 18 August 2026, Gemini in Chrome reached every US Android user, with auto browse limited to paying subscribers. The company even changed its engineering rhythm. In March 2026, Chrome moved to a two week release cycle in response to agentic browser competition.
Anthropic chose the extension route. Claude in Chrome became generally available on paid Claude plans on 26 August 2026, and it can act without approval at every step, with a safety classifier checking each action. The bet is that people will not switch browsers, but they will add an agent to the one they have.
Perplexity went all in on being a browser, and made the agent free. Comet’s agentic capabilities are free, while Atlas’s agent mode was limited to paid plans.
Microsoft pushed Edge toward work. On 23 October 2025 it relaunched Copilot Mode as “your AI browser,” adding Actions, multitab context and a Journeys view. Early testing was uneven: agentic features were inconsistent and remained in limited preview.
The pattern is clear. The winners so far are not necessarily the ones with the best new browser. They are the ones who attach the agent to a place users already are.
The Real Fight: Who Owns the Logged In Session
The deepest conflict is not between AI companies. It is between AI agents and the websites they visit.
Get AIguru’s stories in your inbox
Join Medium for free to get updates from this writer.
Amazon sued Perplexity over Comet. In March 2026 a US court granted a preliminary injunction, accepting that the agent entered password protected areas without authorization, even where users had given permission. In plain terms, the judge separated what you allow from what the platform allows.
Then the pendulum swung back. A US appeals court overturned the injunction, ruling Amazon was unlikely to prove a breach of federal or California computer access laws. The court said it was Perplexity’s users, using the agents, who accessed Amazon’s platform, not Perplexity itself. It was the first federal appeals court ruling on whether AI agents acting for users can legally access online platforms.
Do not read too much into it. The ruling is limited to the facts of the case and does not set a broader legal framework for agentic AI. The question of whether your agent counts as you is still open.
The Security Problem Nobody Has Solved
An agent that acts inside your logged in session is also a target. If a malicious page can talk to your agent, it can try to borrow your access.
The evidence is already there:
- Researchers told an AI browser to buy an Apple Watch, and it completed the purchase on a fake Walmart store set up in minutes.
- On August 3, Zenity Labs widened its PleaseFix research from Comet to five named agents: Claude in Chrome, Gemini in Chrome, Comet, Atlas and Copilot in Edge.
- Attackers can use prompt injection to push AI models toward malicious websites, which can lead to fraud or identity theft.
The companies know this. Google says its agent asks for confirmation before some sensitive tasks. But the risk is structural, not a bug one vendor can patch. An agent built to click quickly can also be fooled quickly.
What This Means for You
You do not need to pick a winner. You need a safe way to use whichever agent you try. Here is a simple routine:
- Start with low stakes tasks. Research, summaries and comparisons first. Hold off on payments.
- Keep payment approval manual. Let the agent build the cart, and you confirm the purchase.
- Use a separate browser profile for agent tasks, so it does not inherit every account you own.
- Check the site address yourself before any checkout, because fake stores are the easiest trap.
- Review what the agent did. Treat its output like a coworker’s work, not a finished fact.
- Expect your tools to move. Atlas users learned that a standalone AI app can vanish, so avoid building your workflow around one product.
Frequently Asked Questions
Did OpenAI shut down ChatGPT Atlas? Yes. Atlas stopped working on 9 August 2026, and OpenAI moved its browser based agent features into ChatGPT and Codex.
What is an AI browser agent? It is an assistant inside a browser that can act on your behalf, such as clicking, filling forms and completing tasks, rather than only answering questions about a page.
Which AI browser should I use in 2026? It depends on your existing tools. Chrome users with a Google AI plan can use Gemini’s auto browse, paid Claude users can add Claude in Chrome, and Comet suits people who want a free standalone option. The security habits above matter more than the brand.
Are AI browsers safe for online shopping? Not fully. Researchers have shown agents completing purchases on fake stores and being steered by hidden instructions. Keep the final purchase approval in your hands.
Will AI browsers replace Google Search? Not by themselves. Search is being merged with agents rather than erased, so the more realistic shift is that fewer tasks start with a list of links.
The Takeaway
The browser war is not really about browsers. It is a fight to be the layer that acts for you across the web, and Atlas shows that how the agent reaches you can change overnight. Choose tools by what they can safely do inside your accounts, not by which one is trending this month.
If you found this useful, follow for more practical breakdowns of how AI is changing everyday tools.
