• AI Weekly
  • Posts
  • The A.I Browser Wars Ignited This Week

The A.I Browser Wars Ignited This Week

Author

Joshua Clear
October 24, 2025

In partnership with

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

The Browser Wars Are Back—And This Time, They're Actually Interesting

Listen, I need to tell you about something wild happening right now in tech. We're in the middle of a genuine browser war for the first time since like, 2008. And I'm not talking about some boring "Chrome versus Firefox" debate your uncle has at Thanksgiving. I'm talking about a fundamental rethink of what a browser even is.

Three major players just dropped AI-powered browsers in the span of three weeks this October: OpenAI launched ChatGPT Atlas, Perplexity made their Comet browser free to everyone, and Microsoft rolled out Edge Copilot Mode. And here's the kicker—they're not just slapping AI into a sidebar and calling it innovation. They're trying to replace the entire search-and-click paradigm with something genuinely different.

The thing is, this could either be the biggest shift in how we use the internet since Google, or it could be a massive security nightmare that makes us all nostalgic for the simple days when browsers just... showed you websites.

Let's break it down.

What the Hell Is an AI Browser Anyway?

Okay, so traditional browsers are basically just windows to the internet, right? You type something into Google, click a blue link, read a thing, click another link, copy some text, paste it somewhere else, open fourteen tabs you'll never close, and eventually give up and just ask your group chat.

AI browsers flip this model. Instead of you navigating the web, the AI does it for you. You tell it what you want—"make dinner reservations for Friday" or "find me the cheapest flight to Tokyo"—and it goes off and actually does the thing. Opens tabs, fills out forms, compares prices, clicks buttons. It's like having an intern who never sleeps and doesn't judge your 3am online shopping habits.

All three of these browsers are built on Chromium (the same engine Chrome uses), so they've got that familiar feel. But they've bolted AI assistants directly into the browsing experience. And not in that annoying Clippy way—these things have actual agency.

ChatGPT Atlas: OpenAI Goes Full Browser

OpenAI dropped Atlas on October 21st, and it's currently only available on Mac (Windows/iOS/Android "coming soon" in that vague tech company way). It's free to use for all ChatGPT users, but the really cool stuff requires paying $20-$200 a month.

The flagship feature is Agent Mode—and this is where shit gets interesting.

Agent Mode lets ChatGPT browse the web autonomously on your behalf. Need to book a hotel? The AI will navigate booking sites, compare prices, fill out your details, and complete the reservation. Want to shop across multiple websites? It'll add items to carts, apply coupon codes, and get you to checkout.

In testing, it worked... kind of. One comparison found Atlas took roughly eight times longer than Perplexity's browser and failed to complete all the requested tasks. Free tier users got told the AI "couldn't perform" actions and instead received helpful suggestions (read: useless). After upgrading to the $20/month Plus tier, things improved, but it was still slower than just doing it yourself.

The sidebar feature is more consistently useful—you can ask it questions about any webpage, summarize content, compare products across tabs, or rewrite text in any field on any website. It's got memory too, so it remembers facts from sites you've visited. Kind of creepy, kind of useful.

Here's why this matters: OpenAI isn't trying to compete with Chrome feature-for-feature. They're betting that conversational interfaces will eventually replace traditional browsing entirely. Why click through fifteen links when you can just ask a question and get the answer?

Perplexity Comet: The Search Company's Power Move

Perplexity launched Comet back in July for $200/month subscribers only. Then on October 2nd, they said "fuck it" and made it free worldwide. This is a pretty aggressive play against both Google Chrome (71% market share) and the new OpenAI browser.

Comet's big differentiator is speed and parallel processing. In head-to-head tests, it completed shopping tasks across three websites simultaneously in under a minute by deploying three parallel agents. ChatGPT Atlas tried to do things sequentially and took eight times longer while only completing two of three items.

The browser has three tiers:

  • Free: Full core features, basic AI models, the works

  • Pro ($20/month): Advanced AI models, 300+ complex queries

  • Max ($200/month): Unlimited everything, plus a "Background Assistant" that handles tasks asynchronously while you do other stuff

That Background Assistant is genuinely novel—you can assign it multiple tasks and go offline, and it'll just keep working. Need research on five different topics? Hand it off and check back in an hour.

But here's where Perplexity gets really interesting: Comet Plus.

They built a revenue-sharing model with publishers. For $5/month (or free if you're already paying for Pro/Max), you get access to paywalled content from CNN, Washington Post, Fortune, LA Times, and others. But here's the clever bit—publishers get 80% of subscription revenue based on how much their content gets used.

Perplexity is trying to solve the "AI is killing media" problem by literally paying publishers. They put up an initial $42.5 million pool and structured it so publishers make money from three types of traffic: direct visits, citations in AI answers, and task completion when the AI uses their content.

This could be brilliant or a total flop, depending on whether it scales. But at least they're trying to build a sustainable model instead of just scraping everything and hoping they don't get sued. (Narrator: They're still getting sued. BBC, Dow Jones, and The New York Times all have active lawsuits against them.)

Microsoft Edge Copilot Mode: The Enterprise Play

Microsoft announced Copilot Mode in July but did a big expansion rollout on October 23rd—just two days after OpenAI's launch. The timing is definitely coincidental, Microsoft swears.

Copilot Mode isn't a separate browser—it's a toggle inside the existing Edge browser. Turn it on, and each new tab becomes an AI-powered starting point for search, chat, or navigation.

The feature set is pretty similar to Atlas: multi-tab reasoning, agentic task automation, memory/context retention. But there's one key difference—Copilot Mode is explicitly labeled as "designed for research and evaluation purposes" and comes with prominent disclaimers that "it may make errors."

In testing, it claimed it deleted an email but didn't. Said it sent an email when it only drafted it in Gmail. It did successfully unsubscribe from a mailing list, so... partial credit?

Microsoft's real advantage here isn't the tech—it's distribution. Edge is already installed on every Windows machine. They don't need you to download a new browser; they just need you to flip a switch. For enterprise customers already paying for Microsoft 365, this is an easy upsell.

The feature is free for now (in the US only for advanced stuff), but Microsoft hasn't said when the free period ends or what pricing will look like after. Classic Microsoft move—get you hooked, then figure out monetization.

The Security Nightmare Nobody Wants to Talk About

Okay, here's where things get genuinely scary.

Within hours of Atlas launching, security researchers at Brave documented critical vulnerabilities. The big one: indirect prompt injection attacks.

Here's how it works: Malicious websites can embed hidden instructions that the AI processes as legitimate commands. An attacker hides text in a webpage—like, literally invisible to human eyes but readable by the AI. When you ask your AI browser to summarize the page, it extracts those hidden commands and executes them.

Brave demonstrated a clipboard injection vulnerability where hidden commands in webpage buttons could make the AI overwrite your clipboard with malicious links. You paste something later, boom—you're on a phishing site handing over login credentials and MFA codes.

Perplexity Comet has its own version called "screenshot-based prompt injection." Attackers use faint colors humans can't see but AI systems extract via OCR. Take a screenshot to ask a question about a Reddit post? Congratulations, you just gave the attacker access to your banking site because you're still logged in.

LayerX researchers found a vulnerability they dubbed "CometJacking"—a single malicious link can hijack the AI assistant and turn it into a spy stealing your data.

OpenAI acknowledged these risks explicitly: "Agents are susceptible to hidden malicious instructions... This could lead to stealing data from sites you're logged into or taking actions you didn't intend." They've run "thousands of hours of focused red-teaming" but admit their "safeguards will not stop every attack that emerges."

The thing is: These vulnerabilities are basically unfixable with current AI technology. LLMs can't reliably distinguish between trusted user input and untrusted web content when constructing prompts. The AI sees hidden commands and treats them as legitimate because it doesn't know the difference.

So we've got this fundamental tension: AI browsers need deep access to your accounts, browsing history, and personal data to be useful. But that same access creates massive attack surfaces for bad actors. And the AI itself can be tricked into becoming the weapon.

Why Google Should Be Terrified (But Probably Isn't Yet)

Google Chrome has 71.77% global market share—about 3 billion users. That's an absurd level of dominance. And they've held it for over a decade.

But here's what's changing: LLM-based search now accounts for 5.6% of desktop search traffic in the US as of June 2025, up from about 2.8% a year earlier. Still small, but growing fast. ChatGPT hit 100 million users in just two months.

More importantly, search behavior is shifting. Click-through rates dropped 30% year-over-year even as search impressions jumped 49%. People are getting answers without clicking through to source websites. When Google's own AI Overviews appear in search results, click-through rates for the #1 position drop 34.5%.

Gartner predicts traditional search engine volume will drop 25% by 2026 as users turn to AI assistants. Analysts project LLM-powered search could surpass traditional search in daily usage around 2029-2030.

Now, Google still processes 15+ billion searches per day versus ChatGPT's tens of millions. They're not dying tomorrow. But the trajectory is clear—conversational AI is eating into their core business.

And here's the deeper issue: Google's business model is fundamentally incompatible with AI answers. They make 75% of Alphabet's revenue from advertising. Ads work because you click through to websites where more ads live. If the AI just gives you the answer directly, there's nothing to click. No clicks, no ads. No ads, no revenue.

Chrome is free because it funnels everyone to Google Search, which prints money. If AI browsers replace that funnel, Google has a real problem.

The Real Question: Will Anyone Actually Switch?

Here's the uncomfortable truth: Probably not. At least not yet.

Chrome has 3 billion users with established workflows, bookmarks, saved passwords, and browser-specific optimizations. The switching costs are enormous. Most people don't even change their default settings—they just use whatever came installed.

And the AI features? They're not reliable enough yet. Reviews consistently describe them as "inconsistent," "slow," and "make errors." When Atlas takes eight times longer to complete a shopping task than doing it manually, why bother?

Plus, there's the trust issue. People are already suspicious of AI. Asking them to give a browser deep access to their banking, email, work systems, and personal data requires an unprecedented level of trust. Especially when security researchers are demonstrating active exploits.

The pricing is also all over the place. OpenAI wants $20-$200/month for full features. Perplexity has the same range. Opera Neon charges $19.90/month. These are significant subscription costs in a market dominated by free browsers.

For comparison: People lost their minds when Netflix raised prices to $15.49/month. Now we're asking them to pay up to $200/month for a browser? Good luck with that.

What Happens Next

The AI browser market in late 2025 feels like the early dot-com era. Lots of innovation, massive capital investment, wildly different business models, and nobody knows what'll actually work.

My prediction? We end up with multiple browser workflows depending on context:

  • Chrome for sensitive stuff (banking, healthcare, anything requiring trust)

  • AI browsers for research and casual browsing

  • Work-specific browsers for enterprise (probably Edge with Copilot)

The winner will be whoever solves the "trust equation" first—delivering genuinely useful AI automation while maintaining security, privacy, and reliability at a level that justifies switching from Chrome.

Right now? Nobody's cracked that code. Atlas is slow and buggy. Comet has active security vulnerabilities. Edge Copilot explicitly disclaims accuracy.

But here's the thing: This is just the beginning. These browsers launched like three weeks ago. The tech will get better. The agents will get faster. The security will (hopefully) improve.

And once one of them figures it out—once there's an AI browser that's demonstrably better than manual browsing—the floodgates open. Because at that point, sticking with Chrome becomes actively stupid.

We're not there yet. But we're getting close.

The browser wars are back, baby! And unlike the boring Chrome/Firefox/Safari debates of the 2010s, this actually matters. Because we're not just arguing about rendering engines and JavaScript performance—we're rethinking the fundamental interface between humans and the internet.

Few other notes:

Opera Neon launched in September for $19.90/month with three separate AI modes (Chat, Do, Make), and reviews describe it as "confusing" because nobody knows which AI to use for what. Classic Opera.

Dia (from the Arc browser folks) is doing an invite-only beta that's less "agentic" but better at workflow management. It's trying to be "AI-first" rather than "AI-added-on," which is a meaningful distinction.

Perplexity tried to buy Chrome for $34.5 billion in August just to flex. OpenAI testified they'd be interested if a judge forced Google to sell it (the judge didn't).

The whole publisher revenue-sharing model Perplexity built is genuinely fascinating as an economic experiment. If it works, it could be the template for how AI companies compensate content creators. If it fails, expect more lawsuits.

Oh, and remember: All of this is happening while Google is actively being sued for monopolistic practices, OpenAI is burning billions in compute costs, and everyone's racing to AGI. The browser wars are just the sideshow.

But what a sideshow it is.

What browser are you using right now? Probably Chrome, right? Yeah, me too.

Sources


OpenAI ChatGPT Atlas

Perplexity Comet

Microsoft Edge Copilot Mode

Market Analysis, Security, Benchmarks

Reply

or to participate.