What Is Chrome Auto-Browse? Google's New Agentic Browsing Explained
15 min read

What Is Chrome Auto-Browse? Google's New Agentic Browsing Explained

By the Kyte Tools team Reviewed

On this page

Chrome auto browse is a new feature in Gemini-in-Chrome that lets Google’s AI complete multi-step tasks on the web for you, instead of you clicking through every page yourself. You describe a goal in plain English, and the browser navigates, types, and fills in forms on its own until the job is done. Google began rolling it out to Android at the end of June 2026, and it is one of the clearest signs yet that the browser is becoming something you delegate to, not just something you use (Google, 2026). This guide explains what auto browse is, what it can actually do, how to turn it on, and the privacy trade-offs you should weigh first.

Key Takeaways

  • Auto browse is an agentic feature of Gemini in Chrome, powered by Gemini 3. It performs multi-step web tasks for you rather than just answering questions.
  • It started reaching US Android devices at the end of June 2026, gated to AI Pro and AI Ultra subscribers, on phones running Android 12+ with at least 4GB of RAM.
  • It can research prices across dates, fill forms, book parking, schedule appointments, and add items to a shopping cart, then pause for you to approve anything sensitive.
  • The privacy story is the real story. Auto browse can touch your Gmail, Calendar, and saved passwords, which makes oversight and tight scoping the whole game.

The browser has been the same basic tool for thirty years: you open tabs, you click links, you do the work. Auto browse is Google’s bet that the next version of the browser does the clicking for you. Below, we break down how it works, where it is showing up, and what to watch for before you let it loose on your accounts.

What is Chrome auto browse?

Auto browse is the agentic mode of Gemini in Chrome. “Agentic” means the AI does not just respond to you; it takes a sequence of actions in the real world to reach a goal. You give Chrome a task, and Gemini plans the steps, opens and reads pages, clicks buttons, fills in fields, and keeps going until it either finishes or needs your input. Google frames it as turning Chrome “from a tool you use to browse the web into a partner that can actually perform tasks on your behalf” (Google, 2026).

The engine underneath is Gemini 3, Google’s flagship model. That matters because agentic browsing is hard: the AI has to understand a messy, ever-changing web page, decide what to click, and recover when something does not go as planned. The feature first appeared as a desktop preview in the US in January 2026 and then expanded to Android at the end of June (Google, 2026).

If the idea feels familiar, that is because Google has been building toward it for more than a year. Auto browse is the descendant of Project Mariner, an experimental DeepMind browser agent first shown in December 2024 that could juggle around ten tasks at once. Google wound Mariner down in 2026 and folded its technology into shipping products like Gemini in Chrome (Wikipedia, 2026). Auto browse is what that research looks like once it leaves the lab and lands in a browser that billions of people already open every day.

Our read: The simplest way to grasp auto browse is by the verb. Old Chrome answers “show me.” Auto browse answers “do it.” When the AI’s job ends in “book it,” “fill it,” or “compare and add to cart,” you have crossed from a search box into an agent. That single shift is why the feature is both genuinely useful and worth approaching carefully.

Where it is rolling out, and on which phones

Auto browse is arriving on two tracks, and it helps to keep them separate. On desktop, it launched first, as a US preview in January 2026 for Google AI Pro and AI Ultra subscribers. On Android, it began rolling out at the end of June 2026 (Google, 2026).

The Android requirements are specific. Google says the feature is reaching select devices in the US running Android 12 or higher, with at least 4GB of RAM and the language set to English-US, and that auto browse itself is limited to AI Pro and AI Ultra subscribers (Google, 2026). In other words, it is a paid, US-first, English-first launch, not a global switch-flip.

What about the headline that this is landing on the Galaxy S26 and Pixel 10? That framing comes from the tech press covering the rollout, not from Google’s own announcement, which names no specific phones (Beebom, 2026). It is a reasonable read, because Google’s latest flagships comfortably clear the Android 12 and 4GB bar, and a related Gemini feature that automates other apps did begin on the Galaxy S26 earlier in 2026 (9to5Google, 2026). Just treat “S26 and Pixel 10 first” as informed reporting rather than an official device list.

The reason any of this matters is reach. Auto browse is not bolted onto a niche app; it is being built into the browser that runs most of the web.

Global browser market share (May 2026, % of users) Chrome 70.3% Safari 15.7% Edge 5.1% Firefox 2.2% Samsung 1.9%
Source: StatCounter Global Stats, May 2026.

Chrome holds roughly 70% of the global browser market (StatCounter, 2026). When agentic features ship inside a browser at that scale, even a slow, gated rollout puts autonomous browsing in front of more people than any standalone AI agent has ever reached.

What can auto browse actually do? Five real tasks

The promise is broad, so it helps to anchor on what Google has actually demonstrated. The pattern across the examples is consistent: auto browse is best at narrow, multi-step chores that involve hopping between pages and copying details around. Here are five jobs it is built for.

1. Compare prices and options across many pages

The flagship demo is research that would normally mean a dozen open tabs. You can ask auto browse to check hotel and flight costs across a range of dates, or to gather quotes from several contractors, and it visits the sites, pulls the numbers, and assembles them for you (Google, 2026). Instead of you running the same search fifteen times, the agent does the legwork and reports back.

2. Fill out forms and handle paperwork

Form-filling is ideal agent territory because it is tedious and rule-based. Google has shown auto browse completing online forms, scheduling appointments, renewing a driving license, gathering tax documents, and filing an expense report (Google, 2026). These are the chores people put off for weeks, and they are exactly the kind of bounded task an agent handles well.

3. Shop visually and build a cart

Auto browse can turn a picture into a purchase-in-progress. In Google’s example, you point it at an item, and it identifies the product, finds it online, adds it to your cart, and even applies a discount code (Google, 2026). Crucially, it stops before paying. The related app-automation feature on Galaxy phones works the same way: it will load up your order but leave the final checkout to you (9to5Google, 2026).

4. Manage subscriptions and recurring orders

The agent is good at small account-management jobs that are easy to forget. Google highlights using auto browse to verify that bills have been paid, manage subscriptions, and update a recurring order, like topping up your pet food on a retailer’s site (Google, 2026). These are low-stakes, repetitive tasks where a few minutes saved each time adds up.

5. Run quick real-world errands

On Android, the pitch leans into everyday errands: reserving parking through a booking site, or copying a grocery list out of your notes and into an online cart (Google, 2026). The common thread is that the agent connects something you already have, like a note or a goal, to a website that can act on it.

Our read: Notice what every one of these tasks has in common. Each is narrow, repetitive, and ends in a clear finish line, and each one stops short of the truly irreversible step. That is not an accident. Auto browse is designed to do all the boring middle work and then hand you the wheel for the final decision. If you want to understand the broader shift this represents, our beginner’s guide to agentic AI covers how to scope these agents safely.

How to turn on auto browse and use it

There are two prerequisites before auto browse will work for you: the right subscription and the right device. You need a Google AI Pro or AI Ultra plan, and, on mobile, a US device running Android 12 or higher with at least 4GB of RAM, set to English-US (Google, 2026). On desktop, the same subscription requirement applies for the US preview.

Once you have access, using it is mostly a matter of asking. You open Gemini inside Chrome and describe the task in plain language, the same way you would type a message to any chatbot. The difference is what happens next: rather than writing you an answer, Gemini begins working through the steps in the browser, and it pauses to ask for confirmation before anything sensitive (Google, 2026).

To make the agent genuinely useful, you can connect it to your personal context. Google lets you opt in to integrations with apps like Calendar, Keep, and Gmail through Gemini’s Connected Apps settings, so the agent can, for example, pull a date from your calendar or a list from your notes (Google, 2026). This is opt-in by design, and it is also the part that deserves the most thought, which is where privacy comes in. The exact menus and buttons will shift as the feature matures and reaches more Chrome versions, so treat the activation flow as a moving target rather than a fixed recipe.

The privacy and safety concerns

Here is the trade-off at the heart of auto browse. To act for you, the agent needs access to the things you act with: your logins, your saved details, and increasingly your email and calendar. When you turn on the integrations, Gemini can read from connected apps like Gmail and Calendar, and to fill forms it can draw on the names, addresses, and credentials saved in Chrome’s password manager (Techlicious, 2026). That is a lot of trust to hand a piece of software, and it is reasonable to pause before you do.

Google has built real guardrails, and they are worth knowing. The company describes several named defenses: a “User Alignment Critic,” a separate AI model that checks whether a proposed action actually matches your stated goal; “Agent Origin Sets,” which restrict the agent to acting only on websites relevant to the task; a work log that records what the agent does; and a dedicated prompt-injection classifier that runs alongside the agent’s planning (SecurityWeek, 2025). On top of that, the agent is required to stop and get your explicit approval before high-stakes actions like making a purchase, signing in, or posting to social media (Google, 2026).

The biggest technical risk is something called prompt injection: a malicious web page or email hiding instructions that try to hijack the agent into doing something you never asked for. This is not theoretical. Google reported a 32% relative increase in malicious prompt-injection activity between November 2025 and February 2026, and security researchers have demonstrated hidden instructions tucked inside calendar invites that could sway an AI with calendar access (Google, 2026).

The threat is growing +32% rise in malicious prompt-injection activity, Nov 2025 to Feb 2026
Source: Google Security Blog, 2026.

Independent voices are cautiously skeptical. Writing at Techlicious, Sean Captain noted he was less worried about Google’s stated protections than about how much sensitive information Gemini could absorb simply by being allowed to read your Gmail, and he recommended using auto browse for low-stakes tasks rather than financial accounts (Techlicious, 2026). The broader security field agrees on the shape of the danger: indirect prompt injection is widely ranked as the top emerging threat to AI applications heading into 2026.

Our read: Every privacy risk here shrinks when you scope the agent tightly. Give it the narrowest task and the fewest connected apps it needs, keep the confirmation prompts on, and start with errands where a mistake costs you minutes, not money. Auto browse is safest treated like a capable but new assistant: useful from day one, trusted with the keys to your accounts only after it has earned it.

How auto browse compares to other agentic browsers

Google is not first to agentic browsing; it is the largest player to make it mainstream. Through 2025 and 2026, a wave of AI browsers and agent modes arrived, each taking a slightly different approach. If you are weighing your options, our comparison of the best AI tools in 2026 goes deeper on the underlying assistants; here is how the agentic-browsing field lines up.

ToolWhat it isNotable angle
Chrome auto browseAgentic mode inside Chrome, powered by Gemini 3Built into the world’s most-used browser; confirmation gates on sensitive steps
ChatGPT (Atlas / Agent Mode)OpenAI’s browser and agent, successor to OperatorStrong at multi-step tasks; available on Plus and Pro tiers
Perplexity CometAI browser focused on research and actionsKnown for sourced, citation-first answers and free availability
Microsoft Edge CopilotCopilot-driven autonomous tasks in EdgeDeep integration with the Microsoft and Windows ecosystem
Anthropic ClaudeClaude in Chrome extension and desktop controlOften favored for complex, technical, computer-control tasks

The specifics of rival pricing and capabilities shift quickly and much of the detail comes from secondary coverage, so treat the comparisons above as a map rather than a spec sheet (No Hacks, 2026). The strategic point is simpler: with Chrome’s roughly 70% market share, Google can put agentic browsing in front of more users than any competitor, which is why auto browse matters even if rival agents are technically comparable.

The trend is industry-wide, not a Google one-off. Gartner projects that 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% the year before (Gartner, 2025). Auto browse is the consumer-facing edge of that same shift.

Frequently Asked Questions

Is Chrome auto browse free?

Not for the autonomous part. The wider set of Gemini-in-Chrome features is rolling out more broadly, but auto browse itself is limited to Google AI Pro and AI Ultra subscribers in the US at launch (Google, 2026). Those plans start around $20 per month, with a higher Ultra tier above it.

Which phones get Chrome auto browse first?

Google’s official announcement only specifies the requirements: a US device running Android 12 or higher with at least 4GB of RAM, set to English-US, starting at the end of June 2026 (Google, 2026). Tech press coverage reports recent flagships like the Galaxy S26 and Pixel 10 as early devices, but Google’s own post does not name specific phones.

Is Chrome auto browse safe to use?

It is built with guardrails: it pauses and asks for your confirmation before sensitive actions like making a purchase, signing in, or posting to social media, and Google runs a prompt-injection classifier to block malicious instructions hidden in web pages (SecurityWeek, 2025). The honest answer is that it lowers risk rather than removing it, so it is best used for low-stakes tasks at first, not your bank or health accounts.

What can Chrome auto browse actually do?

It completes multi-step tasks across websites on your behalf: researching hotel or flight prices across dates, filling out forms, booking parking, scheduling appointments, comparison shopping and adding items to a cart, and updating recurring orders (Google, 2026). It clicks, types, and navigates the way you would, then hands back control for the final confirmation.

How is auto browse different from regular Chrome or the ChatGPT agent?

Regular Chrome is a tool you operate; auto browse operates the browser for you toward a goal you describe in plain language. It is Google’s answer to agentic browsers like OpenAI’s ChatGPT and Perplexity’s Comet, with the difference that it is built directly into the world’s most-used browser rather than a separate app.

The bottom line

Chrome auto browse is the moment agentic browsing goes mainstream. By building autonomous, multi-step task completion directly into a browser that runs roughly 70% of the web, Google has done what no standalone AI agent could: put a capable digital assistant one click away for a huge share of internet users. The Android rollout that began at the end of June 2026 is the start of that, gated for now to paying subscribers in the US.

The smart way to approach it is the same as with any agent. Start with one narrow, low-stakes task, keep the confirmation prompts switched on, and connect only the apps the job truly needs. Auto browse will happily do the tedious middle of your web chores and hand you the wheel for the decisions that matter. Let it prove itself on small errands first, and widen its leash only as your trust grows.

Sources

Keep reading