Google today introduced Chrome auto browse, a feature that lets an AI agent complete multi-step tasks such as booking travel and filling out forms without ongoing human supervision, while separately expanding Gemini in Chrome to desktop users in the United Kingdom, according to two announcements published on the company's official blog, The Keyword.

The UK rollout, detailed in a post by Charmaine Dsilva, Director of Product Management, brings many of Chrome's existing AI features to desktop users in the country starting today, with an iOS expansion promised for next month. A second, more substantial announcement, authored by Parisa Tabriz, Vice President of Chrome, outlines a broader set of updates for MacOS, Windows, and Chromebook Plus users built on Gemini 3, which Google describes as its most intelligent model. The two posts, published within hours of each other, mark one of the more consequential days for Chrome's AI roadmap since the browser's original Gemini integration in September 2025.

Because both features were announced and rolled out on the same calendar day as this article's publication, the developments described below are current as of writing.

What is changing in the UK today

Until this rollout, Gemini in Chrome had been largely confined to United States desktop users since its original September 2025 debut. Chrome received its largest upgrade in browser history on September 18, 2025, according to prior PPC Land reporting, with Google introducing 10 artificial intelligence features powered by Gemini across desktop and mobile platforms, with the rollout beginning for Mac and Windows desktop users in the United States with English language settings. Today's announcement extends that same assistant, now updated with new capabilities, to a second national market for the first time at this scale.

According to the UK-specific announcement, Gemini in Chrome functions as a personalized browsing assistant capable of summarizing lengthy content and comparing information across multiple open tabs. The integration reaches into several existing Google products: users can schedule meetings through Calendar, check locations via Maps, draft emails in Gmail, and ask questions about YouTube videos, all without navigating away from the page they are viewing. The assistant also retains context from past conversations, according to the announcement, allowing it to produce more tailored answers to questions asked across different browsing sessions. A further capability lets users transform images encountered on the web using a simple text prompt, powered by what Google calls Nano Banana 2 capabilities.

Security received explicit attention in the UK announcement. According to the post, Google's models are trained to recognize known threats, including prompt injection, and the system includes safeguards that ask for confirmation before completing sensitive actions. That framing echoes concerns raised earlier this year about a different category of AI browser. According to Brave's security research team, in a blog post examining vulnerabilities in Comet and other AI browsers, enterprise security analysis found Comet up to 85 percent more vulnerable to phishing and web attacks compared to Chrome or other traditional browsers, prompting recommendations that browsers isolate agentic browsing from regular browsing sessions. Whether Chrome's stated safeguards hold up under similar adversarial testing remains to be independently verified, since the announcement offers no technical detail beyond the general assurance that known threat patterns are recognized.

The side panel and what it replaces

The larger of the two announcements, authored by Tabriz, introduces a redesigned interface for Gemini in Chrome: a persistent side panel that stays open regardless of which tab a user is viewing. Previously, Gemini in Chrome operated through a more limited interaction model. The side panel is designed to support multitasking, letting a user keep primary work open in one tab while handling a separate task, such as comparing options across many open tabs or summarizing product reviews scattered across different sites, in the panel itself.

Google's testers, according to the announcement, used the side panel for tasks including comparing options across too-many-tabs, summarizing product reviews across different sites, and finding time for events across chaotic calendars. No quantitative data accompanied these examples; the announcement describes testing in qualitative terms only, without disclosing the number of testers involved or measured outcomes.

Auto browse: an agentic feature limited to paid US subscribers

The most structurally significant addition is Chrome auto browse, a feature Google describes as extending Chrome's existing autofill technology, which has for years automatically entered addresses or payment details, into what the company calls agentic action. According to the announcement, auto browse is available today only to AI Pro and Ultra subscribers in the United States, a narrower launch than the UK-focused side panel rollout described in the companion post.

Auto browse is pitched as a tool for offloading complex, multi-step chores. The announcement lists researching hotel and flight costs across multiple date combinations, scheduling appointments, filling out online forms, collecting tax documents, obtaining quotes from plumbers and electricians, checking whether bills have been paid, filing expense reports, managing subscriptions, and renewing driving licenses as examples drawn from Google's tester group. A more elaborate example describes a user planning a themed party who shows auto browse an inspiration photo; the tool, using what the announcement describes as the multimodal capabilities of Gemini 3, identifies items in the image, searches for similar products, and adds them to a cart while staying within a stated budget and applying discount codes where available. If granted permission, auto browse can also draw on Google Password Manager credentials to complete tasks that require signing in, according to the post.

Google frames the feature's guardrails in similar terms to the UK security language: auto browse is designed to pause and explicitly request confirmation, or prompt the user to complete certain steps directly, before finalizing actions such as a purchase or a social media post. Whether these confirmation checkpoints withstand adversarial prompt injection, the same vulnerability category documented in Comet, is not addressed with technical specificity in either announcement.

Auto browse and the wider AI browser security debate

Agentic browsing tools that act autonomously on a user's behalf have drawn sustained scrutiny from independent security researchers over the past year. Academic work on the subject has been blunt about the scale of the problem. A benchmark developed by researchers at FAIR at Meta for evaluating web agent security against prompt injection attacks found that even top-tier AI models, including those with advanced reasoning capabilities, can be deceived by simple, low-effort human-written injections in realistic scenarios, with attacks partially succeeding in up to 86 percent of cases. Separate analysis published alongside that research characterized the resulting state of the field, only partly in jest, as security by incompetence, since even successful attacks often failed to fully achieve an attacker's stated goal.

Real-world disclosures have reinforced the theoretical concern. According to Brave researchers Artem Chaikin, Senior Mobile Security Engineer, and Shivan Kaul Sahib, Vice President of Privacy and Security, published findings under the title Unseeable prompt injections in screenshots documented additional vulnerabilities in Comet and other AI browsers, and withheld one further vulnerability found in another browser pending additional disclosure the following week. Separate academic analysis of the same underlying pattern described how Brave's security team demonstrated indirect prompt injection attacks against Perplexity Comet, in which attackers hid adversarial instructions in elements invisible to the user, such as white text on white backgrounds or HTML comments, causing Comet to execute sensitive cross-site actions including fetching one-time passwords from email or accessing banking portals when a user merely asked it to summarize a page.

Neither of the two Google announcements references these prior disclosures, nor do they cite independent third-party security audits of Chrome's own agentic feature set. The stated safeguards, training models to recognize known threats and pausing for confirmation before sensitive actions, are presented without supporting technical documentation, benchmark results, or red-team findings specific to Chrome auto browse.

Universal Commerce Protocol becomes part of the browser layer

A separate but connected development folds into this announcement: Chrome will support Google's Universal Commerce Protocol, described in the Tabriz post as a new open standard for agentic commerce co-developed with industry leaders including Shopify, Etsy, Wayfair, and Target. The announcement states this standard ensures AI agents can take actions on a user's behalf seamlessly in Chrome.

This is not UCP's debut. According to prior PPC Land coverage, Google launched the Universal Commerce Protocol with major retailers on January 11, 2026, introducing checkout APIs and payment standards for autonomous AI agents and establishing open-source technical standards for AI agents to execute purchases across different retail platforms without requiring custom integrations for each merchant. The protocol establishes four participant roles, Platforms, Businesses, Credential Providers, and Payment Service Providers, and specifies core capabilities including checkout session management, identity linking through OAuth 2.0, order lifecycle webhooks, and payment token exchange. Since that January launch, the protocol's governance and reach have both expanded substantially. Amazon, Meta, Microsoft, Salesforce, and Stripe joined the UCP Tech Council on April 24, 2026, joining founding members Google, Shopify, Etsy, Target, and Wayfair. The following month, Google used its Google Marketing Live 2026 event to announce a broad expansion of UCP, extending the agentic checkout infrastructure to hotel booking, local food delivery, three new markets, and a wider set of payment options.

Adoption on the merchant side, however, has lagged the protocol's institutional momentum. A study by Originality.ai scanning more than 3 million websites found just 26 public UCP implementations as of May 2026, with none belonging to the major companies that co-developed or endorsed the standard. Today's confirmation that Chrome itself will support UCP places the protocol directly inside the browser layer for the first time, potentially changing that adoption calculus by embedding the standard in software already running on hundreds of millions of desktops, rather than relying solely on merchant-side implementation.

Personal Intelligence, still months away

A fourth element of the Tabriz announcement describes Personal Intelligence, a feature Google says is already available in the standalone Gemini app and will arrive in Chrome in the coming months. The feature is opt-in, according to the post, and users retain control over which apps are connected and can disconnect them at any time. Google frames the intended outcome as transforming the browsing experience from a general purpose tool into a trusted partner that understands the user and provides relevant, proactive, and context-aware assistance. No firm release date accompanies this element of the announcement, leaving it the least concrete of the changes described.

Why this matters for marketers, publishers, and advertisers

The simultaneous UK launch and Gemini 3 upgrade lands atop a browser market that Chrome already dominates by a wide margin. According to Cloudflare's Q3 2025 Browser Market Share Report, Chrome held 66.282 percent of global web traffic, with Safari at 15.135 percent, Edge at 7.492 percent, and Firefox at 3.764 percent. Any behavioral shift Google introduces at the browser level, particularly one involving autonomous agents completing forms, bookings, and purchases, therefore touches a majority of global web sessions by construction.

That scale is precisely what makes the measurement question so pressing. Agentic sessions are difficult to distinguish from ordinary human browsing in standard analytics tooling. Reporting on a related category of AI browsing agent found that categories such as browser and device actively complicate traffic classification, since one competing browser extension presents in server logs as an ordinary browser with an extension attached, and the practical consequence is that agentic sessions appear in analytics dashboards as browser traffic, often without clear attribution, meaning a product listing viewed 40 times in an hour by a price-comparison agent looks, in standard reporting, identical to 40 separate human visits. Chrome's own auto browse feature, now performing form-filling, booking, and shopping actions autonomously from inside the browser itself rather than through an extension, raises the same attribution question in a more direct form, since the resulting sessions originate from the browser vendor's own infrastructure rather than a third-party layer sitting on top of it.

The commerce dimension compounds this concern for advertisers specifically. If auto browse can search for products, add items to a cart, and apply discount codes autonomously, as described in Google's own example involving party decorations, the customer journey that paid search and shopping campaigns are built to measure begins to blur. A user's stated intent may no longer map cleanly onto a single search query or a single click; instead, an agent may execute several searches, comparisons, and a purchase within one side-panel session, with the underlying advertising impressions and clicks distributed in ways current attribution models were not built to capture.

Publisher-facing consequences follow a similar logic to those already documented around AI Overviews, a different but related Google product. A randomized field experiment involving 1,065 desktop Chrome users, published on the Social Science Research Network on April 3, 2026, and last revised June 17, 2026, found that outbound organic clicks fell by 39.8 percent and zero-click searches rose by 34.5 percent when an AI Overview was shown, while clicks on sponsored results stayed flat. That study concerned search results pages rather than in-browser agentic actions, but the underlying mechanism, an AI system completing an information-gathering or transactional task on the user's behalf without requiring a page visit, is directly analogous to what auto browse and the Gemini side panel are designed to do. If a user asks Gemini in Chrome to summarize product reviews across several sites rather than visiting each one, the publisher hosting those reviews may register the AI's fetch of the page without any corresponding advertising impression or engagement metric a human visit would have generated.

Google has not published data quantifying how the side panel, auto browse, or UCP-enabled checkout affects publisher traffic, advertiser attribution, or search click-through rates. The two announcements describe testing in qualitative terms, citing what testers used the features for without disclosing sample sizes, satisfaction scores, or measured behavioral outcomes. That absence of quantitative disclosure is consistent with the pattern in Google's earlier UCP and Universal Cart announcements, which similarly described scale in terms of product catalog size and processed interactions rather than incremental effects on existing measurement systems.

Timeline

  • September 18, 2025: Chrome introduces its first major Gemini integration, described at the time as the browser's largest upgrade in its history, launching for Mac and Windows desktop users in the United States.
  • January 11, 2026: Google launches the Universal Commerce Protocol at the National Retail Federation conference, co-developed with Shopify, Etsy, Wayfair, Target, and Walmart.
  • April 24, 2026: Amazon, Meta, Microsoft, Salesforce, and Stripe join the UCP Tech Council, expanding governance beyond the protocol's founding members.
  • May 19 to 20, 2026: Google unveils the Universal Cart and further UCP expansion at Google I/O 2026 and Google Marketing Live 2026, extending the protocol to hotel booking and food delivery.
  • May 21, 2026: An Originality.ai study finds only 26 public websites have implemented UCP out of more than 3 million scanned.
  • July 14, 2026: Google announces Chrome auto browse for US-based AI Pro and Ultra subscribers and, in a separate post the same day, expands Gemini in Chrome to UK desktop users while introducing a Gemini 3-powered side panel and UCP support within Chrome itself.

Summary

Who: Google, through Chrome Vice President Parisa Tabriz and Director of Product Management Charmaine Dsilva, announced the changes. The updates affect Chrome desktop users in the United Kingdom, AI Pro and Ultra subscribers in the United States, and merchants including Shopify, Etsy, Wayfair, and Target through the Universal Commerce Protocol.

What: Google introduced Chrome auto browse, an agent that completes multi-step tasks such as form-filling, travel research, and shopping without ongoing supervision, expanded Gemini in Chrome to UK desktop users, launched a persistent side panel interface built on Gemini 3, and confirmed Chrome will support the Universal Commerce Protocol for agentic checkout.

When: Both announcements were published today, July 14, 2026, with the UK rollout described as beginning today and an iOS expansion promised for next month.

Where: The UK rollout applies to desktop users in the United Kingdom. Auto browse is limited to AI Pro and Ultra subscribers in the United States. The side panel and Nano Banana image transformation features apply more broadly across MacOS, Windows, and Chromebook Plus.

Why: The announcement matters because Chrome commands roughly two-thirds of global browser traffic, meaning any change to how users search, click, and transact inside the browser has outsized consequences for advertisers, publishers, and merchants. Autonomous agentic features complicate existing attribution and analytics models, while independent security research has repeatedly identified prompt injection as a serious, largely unresolved vulnerability in comparable AI browsing tools, a context absent from Google's own announcements.