Google today expanded Gemini Spark, its background AI agent, to Google AI Pro subscribers in more than 160 additional countries while separately giving the tool the ability to browse Chrome directly using a person's saved logins and passwords, according to a post published on the Google Keyword blog.

The announcement, authored jointly by Adam Coimbra, Director of Product Management for the Gemini app, and Charmaine Dsilva, Senior Director of Product Management, describes two changes rolling out simultaneously: a new feature called Chrome auto browse, and a geographic expansion that brings Spark's automation capabilities to a far larger share of Google's global subscriber base than it previously reached. Neither change is available in the European Economic Area, Nigeria, Switzerland, or the United Kingdom, according to Google's own support documentation for the feature.

What Chrome auto browse does

Gemini Spark first launched at Google I/O 2026 on May 19 as a personalized background assistant built on the Gemini 3.5 Flash model, connecting to Workspace tools and, later, to a handful of third-party services through the Model Context Protocol. Today's update pushes that assistant further into the browser layer itself. According to the announcement, Spark can now integrate directly with Chrome to complete errands that previously required manual, multi-step navigation across several websites.

With a person's permission, the tool can draw on logged-in accounts and saved passwords stored in Chrome to handle what Google describes as tedious tasks: scheduling viewings for apartments a person has already saved, or researching flight options and beginning the booking process on their behalf. Google frames these examples as representative rather than exhaustive, describing the underlying capability as extending Spark's existing web browsing function rather than introducing an entirely new one.

Two safeguards accompany the feature, according to the announcement. The first addresses prompt injection, a well-documented attack vector in which malicious instructions are embedded in content an AI agent processes, such as a webpage or document, causing the agent to take unintended actions. Google states that Chrome auto browse includes protections against this threat, though the announcement provides no technical detail about how those protections function or whether they have been tested against known injection techniques. The second safeguard concerns sensitive actions specifically. Rather than allowing Spark to complete a payment autonomously, the system is designed to hand the task back to the person at that stage, according to the post.

Geographic and access limits

Chrome auto browse is not launching everywhere at once. According to the announcement, the capability is initially rolling out in the United States only, with Google stating plans to expand to additional regions in the future without specifying a timeline. Separate support documentation for Gemini Spark lists further conditions that apply regardless of geography: a person must be 18 or older, must sign in using a personal Google Account rather than a work or school account, must hold a Google AI Pro or Ultra subscription, and must have Keep Activity enabled. The same documentation states that Spark is currently unavailable in the European Economic Area, Nigeria, Switzerland, and the United Kingdom, and that it functions only through the Gemini mobile app, the Gemini app on Mac, and the web app at gemini.google.com.

The auto browse capability itself carries an additional restriction beyond Spark's general availability. Google's support materials specify that, at present, only Gemini Spark running in desktop Chrome can use auto browse, meaning the feature does not yet extend to mobile browsing sessions even where Spark itself is accessible.

The 160-country expansion

Distinct from the Chrome integration, Google is also widening who can reach Spark at all. According to the announcement, access is opening today for Google AI Pro subscribers in more than 160 additional countries, a figure the company presents without naming the specific markets involved. Google AI Pro is priced at 19.99 US dollars per month, according to PPC Land's earlier reporting on the plan's structure.

This expansion sits inside a broader pattern Google has followed with Spark since its introduction. The tool has repeatedly reached new platforms or markets in stages tied to subscription tier, rather than opening fully on day one. When Spark arrived on macOS on June 30, 2026, the beta was restricted to Google AI Ultra subscribers aged 18 and over, and only in the United States at launch, according to Google's announcement at the time, a considerably narrower release than today's country expansion. Ultra remains Google's most expensive consumer tier, priced at 99.99 US dollars per month. Whether today's 160-country rollout for Pro subscribers will eventually be matched by a similar expansion for the macOS build, or by an extension of Chrome auto browse itself beyond the United States, is not addressed in either the announcement or the accompanying support documentation.

How this fits Spark's broader rollout

Gemini Spark launched as a personalized background assistant integrating with Workspace tools and MCP at Google I/O on May 19, 2026, connecting initially to Canva, OpenTable, and Instacart, with that release limited to the United States, as PPC Land reported at the time. Marie Haynes, an SEO consultant quoted in that coverage, described Spark then as marking a shift for Gemini "from an assistant that can answer your questions into an active partner that does the real work on your behalf and under your direction."

The following month, on June 30, 2026, Google brought Spark to macOS in beta, restricted to Ultra subscribers in the United States, while also roughly doubling the tool's connected-app roster to include Dropbox and Zillow Rentals alongside Google Tasks and Keep. That expansion added five services in total and introduced a feature that tracks topics in real time, extending Spark's automation from chat-based prompts onto desktop file management, such as sorting PDFs saved to a Downloads folder into designated locations.

Chrome itself has been a parallel front in Google's agentic push, separate from the Gemini Spark product line specifically. On July 14, 2026, Google introduced Chrome auto browse as a distinct, Chrome-native feature for AI Pro and Ultra subscribers in the United States, built on Gemini 3, and described as extending the browser's existing autofill technology into what the company termed agentic action. That earlier feature, announced the same day Google expanded Gemini in Chrome to United Kingdom desktop users and introduced a persistent side panel interface, listed examples including researching hotel and flight costs, filing expense reports, and renewing driving licenses. Today's Spark-specific auto browse announcement uses closely related terminology and a similar permission model, though Google's post does not clarify whether the two features share underlying code or represent separate implementations converging on the same name.

Security context Google's announcement omits

Neither today's post nor the Chrome-native version from July addresses independent security research that has scrutinized this category of tool closely. Brave's security research team documented multiple vulnerabilities in Perplexity's Comet browser and other AI browsers, finding that attackers could hide adversarial instructions in elements invisible to users, such as white text on white backgrounds or HTML comments, causing an agent to execute sensitive cross-site actions including retrieving one-time passwords from email or accessing banking portals when a person had merely asked it to summarize a page.

A benchmark developed by researchers at FAIR at Meta, cited in PPC Land's coverage of the July Chrome feature, found that even top-tier AI models could be deceived by simple, low-effort, human-written prompt injections in realistic scenarios, with attacks partially succeeding in up to 86 percent of cases tested. Whether Chrome auto browse's stated protections against prompt injection, mentioned in today's Google post without technical elaboration, would hold up against comparable adversarial testing has not been independently verified.

Why this matters for advertisers and publishers

Gemini Spark's expansion into Chrome-based task automation lands at a moment when the marketing industry is still working out how to measure agentic browsing sessions at all. When Google introduced its Chrome-native auto browse feature in July, PPC Land reported that agentic sessions are difficult to distinguish from ordinary human browsing in standard analytics tooling, since a product listing viewed dozens of times in an hour by an automated agent can appear in server logs identical to the same number of separate human visits. Today's expansion of a comparable capability under the Gemini Spark name, now reaching a substantially larger international subscriber base, extends that same measurement gap to whichever new markets gain access, once Chrome auto browse itself expands beyond the United States.

The commerce implications are direct for paid search and shopping advertisers specifically. If an agent can research flight options, begin a booking, and schedule a property viewing using saved credentials without a person navigating each page manually, the customer journey that attribution models are built to capture starts to blur at exactly the point where intent signals, such as a search query or an ad click, would normally be recorded. Google's own examples in today's announcement, apartment viewings and flight research, both sit inside categories that publishers and travel advertisers have historically depended on organic and paid search traffic to fulfil.

Publishers face a related question. An agent that logs into a site using saved credentials and completes a scheduling form may register as a page fetch without generating the engagement signals, page views, session duration, ad impressions, that a human visitor browsing manually would produce. Google has not published data quantifying how many Spark sessions currently use Chrome auto browse, nor has it disclosed engagement or conversion metrics tied to the feature's earlier United States rollout under the Chrome-native banner. Both today's announcement and the July post describe testing and capability in qualitative terms only.

The regional exclusions carry their own significance for the European market specifically. Spark's absence from the European Economic Area, Switzerland, and the United Kingdom means marketers operating in some of the largest advertising markets outside the United States remain unable to test or plan around the tool, even as it reaches a wide swath of smaller markets elsewhere through today's 160-country expansion. Google's own regulatory disclosures elsewhere have described feature delays in the region tied to compliance obligations under EU frameworks, though today's announcement does not state a reason for Spark's specific exclusion from the EEA, Switzerland, or the UK, nor does it offer a timeline for when, or whether, that exclusion might lift.

Timeline

  • May 19, 2026: Gemini Spark launches at Google I/O 2026 as a background assistant built on Gemini 3.5 Flash, connecting to Workspace tools and, via MCP, to Canva, OpenTable, and Instacart, limited to the United States.
  • June 30, 2026: Google extends Gemini Spark to macOS in beta, restricted to Google AI Ultra subscribers aged 18 and over in the United States, while adding five connected apps including Dropbox and Zillow Rentals.
  • July 14, 2026: Google introduces Chrome auto browse as a Chrome-native feature for US-based AI Pro and Ultra subscribers, built on Gemini 3, alongside a UK expansion of Gemini in Chrome and a redesigned side panel interface.
  • July 30, 2026: Google integrates Gemini Spark directly with Chrome auto browse, using saved passwords and logged-in accounts to complete tasks such as apartment viewings and flight bookings, and expands Google AI Pro subscriber access to more than 160 additional countries, excluding the European Economic Area, Nigeria, Switzerland, and the United Kingdom.

Summary

Who: Google, through Adam Coimbra, Director of Product Management for the Gemini app, and Charmaine Dsilva, Senior Director of Product Management, announced the changes. The update affects Google AI Pro and Ultra subscribers using Gemini Spark, with the geographic expansion specifically reaching Pro subscribers in more than 160 additional countries.

What: Google integrated Gemini Spark directly with Chrome auto browse, allowing the tool to use saved passwords and logged-in accounts to complete tasks such as scheduling apartment viewings and researching flight bookings, with protections stated against prompt injection and a design that hands sensitive actions like payments back to the person. Separately, Google expanded Google AI Pro subscriber access to Spark in more than 160 additional countries.

When: Both changes were announced and began rolling out today, July 30, 2026.

Where: Chrome auto browse is initially available only in the United States, with Google stating plans to expand to additional regions in the future. The country expansion excludes the European Economic Area, Nigeria, Switzerland, and the United Kingdom, according to Google's support documentation for Gemini Spark.

Why: The announcement matters because it extends agentic browsing, an AI system acting autonomously using a person's stored credentials, into everyday commercial tasks like travel booking and property viewings, categories that paid search and travel advertisers have historically depended on measurable search traffic to fulfil. Independent security research has repeatedly identified prompt injection as a serious vulnerability in comparable AI browsing tools, a context absent from Google's own announcement, while the tool's continued exclusion from several major European markets leaves advertisers there unable to test or plan around a capability already reaching a substantially wider international footprint elsewhere.