Google on October 20, 2026 introduced a second, lower-confidence tier of site warnings in Chrome for users of Safe Browsing's Enhanced Protection mode, moved its Gemini-based scam detection from the device to Google's servers, and began inspecting commands that users copy from web pages for signs of the attack technique known as ClickFix. The changes were described in a post on Google's blog titled "Our latest defenses against evolving web scams," written by Jasika Bawa, Group Product Manager for Chrome, and filed under Security and Chrome Security.

The post presents all three as features added during 2026 to Enhanced Protection, the opt-in mode Google launched in 2020. According to Google, "more than 1 billion Chrome users" had it enabled when the company last published a count, in 2025. The post gives no newer figure.

In Short

Chrome can now show a softer warning on websites that look like scams but that Google is not sure about, in addition to the big red warning it already shows on sites it knows are dangerous. This matters to anyone running a website, because a page that only looks deceptive can now get a warning, and to Chrome users who turned on the stronger protection mode, because some page text can be sent to Google for a check. Chrome will also warn people when a site tricks them into copying a harmful command to run on their own computer.

Two tiers instead of one

Until now, a Safe Browsing site warning in Chrome meant one thing: the full-page red interstitial that appears before a page known to be malicious. Google is splitting that into two categories when Enhanced Protection is on. According to the post, suspicious sites carry a "lower confidence verdict, possible risk of user harm," while dangerous sites carry a "high confidence verdict, high risk of user harm."

The model is borrowed from downloads. Google cited an earlier post on securing the Chrome downloads experience, published on the Google Security Blog in July 2024, which set out a two-tier taxonomy separating suspicious files from dangerous ones. The site version follows the same logic.

Suspicious-site warnings, according to Google, "notify users when a page exhibits deceptive patterns commonly associated with known malicious sites." That wording matters. The trigger is resemblance to known bad sites, not a confirmed finding that the page itself is malicious. The new warnings look different from the red interstitial, distinguished "by format, iconography, color, and text," according to the post, so that users can "quickly and confidently make the best choice for themselves based on the context."

Dangerous-site warnings are unchanged. They "will continue to display as the full-page red interstitial warning that users already know," Google wrote.

The most consequential sentence in this section concerns volume. "By tailoring the warning UI in this way, we are able to safely expand our coverage to display a higher volume of protective warnings," the post states. Put plainly, a softer warning lets Google flag pages it would previously have left alone because the evidence fell short of the bar for a red screen.

What the post does not say

Google did not publish the thresholds that separate a suspicious verdict from a dangerous one, nor did it list the "deceptive patterns" that trigger the lower tier. There is no figure for how many more warnings Chrome expects to show. The post also does not describe how a site owner would learn that pages had been labelled suspicious, whether such verdicts appear anywhere site owners can see them, or how a mistaken verdict would be reviewed. It gives no Chrome version number, no rollout schedule and no breakdown by platform.

For publishers and advertisers, those gaps are where the practical questions sit. A landing page built around urgency, countdown timers, imitation system dialogs or download buttons that resemble operating system prompts could share surface features with scam pages. Whether such pages fall inside the new tier is not something the post addresses.

Gemini moves from the device to the server

The second change reverses the architecture Google described in 2025. In May of that year, Google said Chrome 137 had added a layer of protection using Gemini Nano, the on-device version of its language model, to spot tech support scams for Enhanced Protection users. According to Google at the time, the model ran locally and was invoked only by specific triggers, such as a page using the keyboard lock API, and its output went to Safe Browsing for a final verdict.

That approach had limits. It ran only "on supported devices," as the October post acknowledges, and it targeted one scam category. PPC Land reported in May 2026 that Chrome writes a roughly 4 GB Gemini Nano model file to users' devices, after privacy researcher Alexander Hanff published his findings. In September 2025, Google had already said it would extend the Gemini Nano protection beyond tech support scams to fake virus and fake giveaway pages.

According to the October 20 post, the on-device feature "proved successful in combating this category of scam, and helped prove the viability of LLM-assisted detection." Google gave no numbers to support that assessment. To reach every Enhanced Protection user and to cover "fake antivirus and browser update scams," the company wrote, "we have transitioned to a more powerful server-side Gemini model."

What the model looks at

The server-side model reads page content and extracts what Google calls key indicators. The post names three: "the brand represented, the intent of the page, and a scam likeliness score." Safe Browsing then combines that output "with domain reputation and other security signals, to generate a final verdict." Gemini, in other words, does not issue the warning. It supplies signals to a larger system that does.

The brand indicator is notable for anyone whose trademark is commonly impersonated. Scam pages for fake antivirus products and fake browser updates typically borrow the look of well-known software companies, and a model that names the brand a page appears to represent can be compared against the domain actually serving it. The post does not explain how that comparison is made.

Moving inference off the device raises an obvious privacy question, and Google addressed it in two sentences. Chrome "minimizes the amount of data shared with the server by only analyzing page text when a site exhibits high-confidence scam indicators," according to the post. And when page content leaves the device, it "is processed ephemerally - meaning it is never logged, retained or used to train models."

That means Chrome performs some local screening first and sends page text only once that screening finds strong signals. What those local signals are, how often they fire, and how much text is sent per page are not specified. The post includes a diagram titled "Chart of how we're using Gemini for scam protection," but does not describe its contents in the text beyond the steps above.

The privacy framing differs from Google's 2025 description of Enhanced Protection in general. In February 2025, Google wrote that data shared in the mode, including URLs, page contents and files, is anonymized when possible, used only for security, and kept only as long as necessary. The October post's "never logged, retained" language applies specifically to the page text sent for Gemini analysis.

ClickFix: an attack that leaves the browser

The third feature targets a technique that sidesteps browser protections entirely. ClickFix, according to the post, "is a social engineering technique that tricks users into manually copying and executing malicious commands on their own computers in order to 'fix' a fake issue." A page shows a fake error, a fake CAPTCHA or a fake update prompt, then instructs the visitor to paste a command into a terminal or the Windows Run dialog.

The command pasted is "often not the command shown on the page," Google wrote. It "almost always invokes Living Off the Land Binaries (LOLBins) - legitimate, pre-installed operating system tools and executables, such as PowerShell, or cmd.exe, that attackers repurpose to carry out malicious actions such as downloading and setting up malware."

The reasoning behind the technique is laid out clearly in the post. "Because these utilities are native to the system and trusted by security software, bad actors use them to blend in and hide their activity, and by shifting the download process out of the browser and into the terminal, attackers completely sidestep existing Chrome and Safe Browsing download protections." Chrome's file download checks, including the deep scans of suspicious files Google has described for Enhanced Protection, never see a payload that PowerShell fetches on its own.

Google has tracked the technique for months. Its June 8, 2026 fraud and scams advisory, written by Laurie Richardson, Vice President of Trust and Safety, mentioned a ClickFix campaign that used fake browser update lures and distributed malware on Google Sites. The October post links to that advisory.

How the clipboard check works

Chrome's response operates at the moment of copying. According to the post, Chrome "now detects when an Enhanced Protection user is tricked into copying suspicious commands." When a potential ClickFix script is copied, "the clipboard payload is scrubbed of PII and sent to Safe Browsing for further analysis, and if a dangerous verdict is returned, Chrome shows a warning."

Three details stand out. First, Chrome is now inspecting clipboard contents generated from web pages, which is a new category of data for Safe Browsing. Second, personally identifiable information is removed before the command leaves the device, though the post does not define what counts as PII in a shell command or how it is detected. Third, the warning appears only on a dangerous verdict, the high-confidence tier, not the new suspicious tier.

Google wrote that the system "is designed to be high precision, in order to protect users in real time while minimizing disruption to legitimate developer activity." That is a recognition of a real tension. Documentation sites, developer tools pages and software installation guides routinely tell readers to copy commands that call PowerShell or download scripts. A detector tuned to catch ClickFix has to distinguish those from attacks, and the post does not say how. No false positive rate was given.

Ten times safer, by one measure

The post opens with its biggest number. Since 2025, Google wrote, additional features have made Enhanced Protection users "ten times safer from phishing, social engineering, notifications abuse, and other scams when compared to Safe Browsing's Standard Protection mode."

A footnote on the post explains the basis: the figure was calculated by comparing the average number of safety warnings shown to Chrome users in Enhanced Protection with the average shown to users in Standard Protection. It is therefore a measure of warnings displayed, not of scams avoided, money saved or infections prevented. A mode built to show more warnings will tend to score well on that measure, and the new suspicious tier, by Google's own description, raises the volume of warnings further.

The figure also represents a large jump from Google's previous claim. In February 2025, when it announced the 1 billion user milestone, Google said Enhanced Protection made users about twice as safe from phishing and other scams as Standard Protection. The October post does not say whether the two figures use the same method, so the change from roughly twice to ten times cannot be read as a measured fivefold improvement.

Why this matters for advertisers and publishers

For the marketing community, Chrome's warning screens sit directly between a click and a landing page. With Enhanced Protection enabled for more than a billion users by Google's count, the rules governing when a warning appears affect a large share of paid and organic traffic. A red interstitial almost always ends a visit. The post does not say how users respond to the softer suspicious warning, and Google gave no click-through data.

The same scams Chrome is targeting often arrive through ads. PPC Land reported on October 9 that an analytics engineer was phished through a Google Ads listing for Google Cloud Console that led to a page on sites.google.com. In August, PPC Land covered the SourTrade campaign, which used programmatic ads in 12 countries to assemble malware inside victims' browsers and showed harmless pages to bots and researchers. These are examples of malvertising, where the advertising system itself carries the attack, and they illustrate why browser-level checks matter even when ad review fails.

Google's own ad enforcement numbers show the scale. Its 2025 Ads Safety Report said Google blocked or removed 8.3 billion ads, including 602 million linked to scams, and suspended 4 million accounts for scam-related violations. Chrome's changes operate after an ad has been served, at the point where a user reaches the destination. Ad review and browser warnings are separate systems, and the October post does not say whether Safe Browsing verdicts feed back into Google Ads enforcement.

Fake antivirus and fake update pages, two categories the server-side model now covers, are standard landing pages in phishing and tech support scam operations. The losses involved are documented. PPC Land reported on October 19 that the FTC told Congress older Americans lost about $126 million to tech support scams in 2025, and in July that the FBI counted about $1.04 billion in tech and customer support fraud losses among complainants aged 60 and older.

There is a second side for site owners. Pages that legitimately carry brand names they do not own, such as reseller pages, comparison sites, software download portals and affiliate pages, are exactly the kind of content a model scoring "the brand represented" against domain reputation will evaluate. Whether such pages are more likely to receive a suspicious verdict is unknown. So is the effect on affiliate and arbitrage sites that rely on aggressive formats. Google has not described any process for site owners to contest a suspicious verdict.

The server-side shift also changes what Google sees. Previously, the large language model analysis ran on the device and only summarized signals went to Safe Browsing. Now, for pages that trip the high-confidence screen, the page text itself goes to Google's servers. Google states that this text is not logged, retained or used for training. That commitment is the company's own and cannot be checked from the post.

What remains unclear

Several practical points are not covered by the October 20 post:

  • Which Chrome version carries the features, and whether they are live on all platforms, including Android and iOS.
  • Whether the suspicious-site tier and the clipboard check are available only to Enhanced Protection users, as the post implies, or will reach Standard Protection.
  • The criteria and thresholds separating suspicious from dangerous verdicts.
  • Whether site owners are notified of suspicious verdicts, and how to dispute them.
  • How accurate the server-side Gemini model and the ClickFix detector are in practice.

The post ends by asking readers to turn on Enhanced Protection in Chrome and in their Google Account. Google framed the update as a response to attackers shifting tactics: "As scammers find new ways to take advantage of people, Chrome and Safe Browsing will continue to develop smarter, real-time defenses to make the web safer for everyone."

Timeline

Summary

Who: Google's Chrome and Safe Browsing teams, in a post by Jasika Bawa, Group Product Manager for Chrome. The changes apply to Chrome users with Enhanced Protection enabled, which Google counted at more than 1 billion in 2025, and indirectly to website owners and advertisers whose pages may be evaluated.

What: Three additions to Enhanced Protection. A new suspicious-site warning tier for lower-confidence verdicts, distinct from the red dangerous-site interstitial. A switch from on-device Gemini Nano to a server-side Gemini model that scores pages on brand, intent and scam likeliness, now covering fake antivirus and browser update scams. And detection of ClickFix commands copied to the clipboard, which are scrubbed of PII and sent to Safe Browsing. Google also claims Enhanced Protection users are ten times safer than Standard Protection users, based on the average number of warnings shown.

When: The post was published on October 20, 2026. Google described the features as introduced in 2026 without giving dates, versions or a rollout schedule.

Where: In the Chrome browser, through Safe Browsing. The post does not specify platforms.

Why: Google says scams are shifting toward techniques that evade existing defenses, such as ClickFix, which moves malware downloads out of the browser and into system tools like PowerShell. The lower-confidence tier lets Google show more warnings than its high-confidence red screens alone allowed, which raises open questions for legitimate sites that resemble scam pages.