Gemini's Auto Browse Reaches India, and the Agentic Browser Gets Real

Chrome users in India can now hand the browser a task and let Gemini carry it out across pages. Indian technology outlets reported in early October that Google had extended Gemini in Chrome, and its Auto Browse feature, to the country. That turns agentic browsing from a demo into something sitting in front of the largest Chrome audience in the world, and it changes what you have to decide about a browser that clicks on your behalf while your session is live.

The sourcing here deserves to be stated plainly rather than buried. Three outlets carried the rollout — Gadgets360, Digit and ETV Bharat — and all three agree on the substance. None of them is Google, and a vendor page announcing the India rollout was not located when the research for this story was compiled. So the fact of the rollout rests on the reporting, and the details below are marked wherever they are less than settled.

What actually changed for a Chrome user in India

The change is that Auto Browse is available in the country. The reporting describes it as a feature that carries out multi-step tasks in the browser on the user's behalf, inside Gemini in Chrome, rather than as a separate product or an extension you install. For anyone in India running a current Chrome build, the practical question is no longer whether the capability exists but whether it has reached their account.

Google's own primary documentation would normally answer the questions that follow from that. What plans include it, what it costs, which interface languages it speaks, and whether it is on by default for every signed-in profile are exactly the things a vendor page settles. In this case it does not, at least not in a form that could be located and read, and the outlets did not agree on those points either. Treat the plan eligibility and the supported-language list as unconfirmed rather than as whatever a screenshot in a summary suggests.

The date is unconfirmed in the same way. The reporting clusters in the first week of October 2026, and the timestamps on the published pieces suggest 7 October, but that is an inference from when the articles appeared rather than a date Google stated. Regional rollouts also tend to arrive in waves, so two people in the same city can legitimately see different behaviour on the same day.

India is the first market where this is a volume question

Every agentic browsing demonstration so far has run in front of an audience that opted into watching it. India is different in kind. It is where Chrome's user base is largest by volume, which means Auto Browse is arriving in front of the biggest single-country population of Chrome users rather than in front of a launch-week audience in a few cities.

Scale changes the kind of thing that goes wrong. An agent that misreads a page and books the wrong seat is an anecdote. The same agent running for tens of millions of sessions produces a category of incident: support queues full of transactions nobody meant to make, and a recovery path that assumes the user can reconstruct what the agent did from a log they never saw. Payment flows, form submissions and account-changing actions are where that stops being hypothetical, because those are the steps where a mistake costs money and cannot always be undone with a back button.

It is also the first serious test of an agentic browser in a market where Google competes with well-established local assistants rather than mainly with itself. Users there have alternatives that are already embedded in phones and messaging apps, and they will judge Auto Browse against those rather than against a blank slate.

Why a regional rollout is the interesting kind

A staged rollout is not one event, and treating it as one is how teams end up confused about whether a feature is live. The reporting describes availability arriving in India; it does not describe a single switch that turned it on for every account at the same moment. Google's pattern for this kind of change is a gradual expansion by account, by build and by region, which means the same organisation can contain users who have the feature and users who do not, with no visible reason for the difference.

That has a mundane but annoying consequence for anyone writing internal guidance. Instructions of the form "open Chrome, click the Gemini icon, choose Auto Browse" fail for a share of readers whose build has not been included yet, and those readers reasonably conclude that they have done something wrong. If you are documenting this for a team, say that availability is rolling out and give people a way to check rather than a fixed path to follow. Any policy decision you make this month should also be written to survive the feature appearing for more users next month, because that is what a rollout does.

The regional dimension adds a second layer. A feature that exists in one country's build and not another's is, from the browser's point of view, the same browser with a different configuration. Support teams that handle both populations need to know which one they are talking to before they start diagnosing, and the quickest way to establish that is to ask whether the option appears at all rather than which steps the user followed.

The part that makes the browser the agent

An agent that runs in Chrome inherits everything Chrome holds. It has the cookies, the saved passwords, the payment methods, the open tabs and the logged-in admin sessions. It is not an integration asking for a scoped token through an OAuth screen that a reviewer can read. It is the session itself, given instructions.

That is the whole reason the feature is useful and the whole reason it is consequential. The value of doing a task in the browser is that the browser is already authenticated to everything the task touches. The same property means the boundary between what the agent may do and what it may not is a policy inside the product rather than a permission granted to a third party.

For a reader who administers machines rather than owns them, that is a governance problem before it is a productivity feature. Your existing controls live at the network and device layer, and they were built to answer questions about extensions, downloads, and which sites a browser may reach. None of them answers whether an enterprise profile's agent may complete a purchase with a stored card. A managed browser policy that turns Auto Browse off is the kind of control that has to exist before the feature is enabled by default, and whether it does is unconfirmed.

There is a specific version of that problem that matters for anyone shipping a web application rather than using one. An agent driving a browser is, to every server it touches, an ordinary browser. Your analytics count its page views as a human session. Your rate limiting sees a client that behaves like a person until it suddenly does not. Your forms receive submissions with realistic values and no human behind them. That is not a reason to block it, and it is a reason to check what your own abuse controls do when a share of your traffic is an agent acting on a real user's behalf with a real session cookie. A support tool that decides a request is suspicious because it was fast will now be making that decision about a legitimate user.

Prompt injection belongs in the same paragraph, because the agent reads pages it did not write. A page can contain text intended for a model rather than a person, and an agent told to complete a task is, by construction, a reader that may follow an instruction it encounters. Nothing in the reporting says how Auto Browse handles that class of input, and a reader should not assume it is solved.

What the India launch means for data handling

India is also a market where an agentic browser will attract attention for what it does with user data, which is the second reason this rollout matters more than its size. A feature that reads the pages you visit, holds your session and takes actions on your behalf is a different data-processing proposition from a browser that renders pages, and it sits in a jurisdiction whose rules on personal data have been taking shape over the same period this feature was developed. How the feature is configured in that market, and what it retains, is exactly the kind of detail that a vendor documentation page would normally state and that this rollout does not appear to have supplied.

For an organisation operating in the region, that is a reason to route the question through whoever owns compliance rather than through whoever owns the browser fleet. The two are usually different people, and the second one frequently enables a feature without knowing that the first one has an opinion about it.

What a reader should do about it

If you are in India and the feature has reached your profile, decide what it is allowed to touch before you let it touch anything. The useful first step is not a sweeping test. Give it a checkout on a site you do not care about, watch the whole run, and see whether the confirmation screen it shows you is enough to reconstruct what it did. If it is not, you have learned the important thing cheaply.

Then separate the accounts. An agent operating inside your session is a reason to stop using one browser profile for everything, if you have not already. Work profiles that carry admin access to production systems should not be the profile where you experiment with an agent that clicks things.

If you run a support desk or an IT estate in the region, the other half of this is the call volume that follows. Assume someone will let the agent do something they now want reversed, and check now whether your recovery process can answer the question "what did it do, and in what order" without asking the user to remember. Auto Browse that records its steps in a place the user can reach is a different support load from Auto Browse that reports success and nothing else.

Finally, treat the details as live. The plan eligibility, the language coverage and the rollout date are all unconfirmed right now, and the primary documentation that would settle them has not shown up. Anyone writing a policy, a training document or a budget line against this feature should date the assumption and check it again, rather than treating a week of news coverage as the settled specification.