Google has begun rolling out the Gemini AI assistant in the Chrome browser for users in the US on Android devices. The feature lets you get a brief summary of an open page's content, answer questions about its text, interact with other services in the Google ecosystem — Calendar, Keep, and Gmail — and generate images via the Nano Banana tool. In effect, the browser is turning from a search tool into an environment where the assistant takes on part of the routine work with information right inside the tab window.
Technical requirements and limitations
To enable the feature, the device must be running Android 12 or a newer version, have at least 4 GB of RAM, have the latest Chrome update installed, the user must be signed in to a Google account, and the system interface must be set to English. Google separately notes that the assistant is unavailable in incognito mode — a logical limitation tied to the privacy policy of private sessions. Thus, at launch the audience remains relatively narrow: English-speaking US users with fairly modern smartphones.
Auto Browse: an agentic mode for AI Pro and Ultra subscribers
Alongside the basic integration, Google opened a preview of the Auto Browse mode for subscribers of the Google AI Pro and Ultra plans. This is no longer just a helper that answers questions, but an agentic mode capable of independently performing multi-step tasks in the browser: for example, finding parking in a given neighborhood or updating an existing order. The key difference is that the assistant acts sequentially, moving between pages and forms. At the same time, before sensitive actions such as a purchase or posting to social media, Gemini asks the user for explicit confirmation, keeping the final say with the human.
Security and the underlying model
All of the listed features are built on the Gemini 3.1 model. Google states that it has implemented protection against prompt injections — attacks in which malicious text on a page tries to "hack" the assistant's instructions and force it to perform unwanted actions. This is critically important precisely for the browser scenario, where the model constantly processes arbitrary content from open websites. Nevertheless, the company does not disclose the full architecture of these protective mechanisms.
Contradictory data
It is important to honestly note the discrepancy between the marketing claims and the technical transparency. On the one hand, Google positions Auto Browse as a mature agentic mode with injection protection and confirmation of sensitive actions. On the other hand, the company does not specify how long the context of a processed page is stored, and does not publish a full list of websites compatible with Auto Browse. For teams planning enterprise deployment, this means open questions about the boundaries of the assistant's permissions and the reproducibility of its actions across different resources. In other words, the actual readiness for production use has not yet been confirmed by public documentation.
Context: not just Android
It is worth noting that the integration of Gemini into Chrome is not limited to smartphones: in parallel, the AI assistant is also appearing on Chromebook Plus laptops, indicating Google's strategy of evenly covering its entire device lineup. For the market, this is a signal that the browser is becoming the main "window" for agentic AI, not just a search bar. Users from other regions, including Russia, do not currently have confirmed access to these features, and the timing of their availability outside the US has not been disclosed.
