Google Chrome will use Gemini to flag scam sites and warn before you paste a ClickFix command

Chrome is adding a second line of defense against scams that start with a web page and end with a victim pasting a command into their own computer. Google says the browser will use its Gemini model to judge whether a page is deceptive, and will warn a person at the moment a suspicious ClickFix command lands on the clipboard. The changes apply to people who have switched on Safe Browsing’s Enhanced Protection mode.

The day in tech, in one email: TLDR is the daily newsletter with links and TLDRs of the most interesting stories in startups, tech and programming.

Get TLDR →

We may earn a commission if you sign up.

The announcement came from Jasika Bawa, a group product manager on the Chrome team, in a security post dated Oct. 20, and BleepingComputer’s Mayank Parmar covered it a day later. Neither account gives a ship date, a Chrome version number or a platform-by-platform schedule, so the right reading is that the features are rolling out in 2026 rather than switching on everywhere at once.

A two-tier warning for sites that look wrong but are not provably bad

The first change is a new label. Chrome already shows a full-page red interstitial when Safe Browsing is highly confident that a site is dangerous. Bawa wrote that Google is now adding a lighter “suspicious site” tier, modeled on the warning system Chrome introduced for downloads in 2024. A suspicious verdict means a lower-confidence call that a page may cause harm, and the warning differs from the dangerous-site version in its format, icons, color and text.

Google’s stated reason for the split is coverage. A browser that can only choose between silence and a red wall tends to stay silent when the evidence is mixed. With a gentler tier, Google says it can show protective warnings on pages that exhibit deceptive patterns but fall short of the confidence threshold, while keeping the red page for the clearest cases. The company also repeats its claim that Enhanced Protection users are ten times safer than Standard Protection users, a figure its footnote ties to the average number of warnings shown to each group.

Server-side Gemini reads the page for brand, intent and scam likelihood

Last year Chrome began using an on-device Gemini model to spot tech support scams on supported hardware. According to Google’s post, that experiment showed language models can do this job, and the company has now moved to a more powerful server-side Gemini model for every Enhanced Protection user. The scope widens as well, to include fake antivirus pages and fake browser update prompts. The model scores a page on the brand it appears to represent, what the page is trying to get the visitor to do, and how likely it is to be a scam.

The score alone is not the verdict. Safe Browsing blends it with domain reputation and other signals before deciding whether to warn. Bawa wrote that Chrome sends page text off the device only when a site already shows high-confidence scam indicators, and that the content is “processed ephemerally,” which Google defines as not logged, not retained and not used to train models. Those are Google’s own assurances, and the post offers no independent audit of them.

Enhanced Protection is an opt-in setting that Chrome introduced in 2020, and Google said in 2025 that more than one billion Chrome users had it enabled. Anyone who wants to check or change the mode can follow Chrome’s Enhanced Protection help page. The Standard Protection default does not get the new server-side checks.

The clipboard check aimed at ClickFix

ClickFix is the more unusual piece. In these attacks a fake error message, CAPTCHA or “verification” step instructs a visitor to copy a snippet and run it in PowerShell or the Windows Command Prompt, which hands the attacker code execution without any file download for Chrome to inspect. Google’s post notes that the command a victim pastes is often different from the one shown on the page, and that attackers lean on built-in utilities known as living-off-the-land binaries. Google had already flagged the technique in its June 2026 fraud advisory.

Under the new design, when an Enhanced Protection user copies something Chrome considers a suspicious command, the browser removes personal information from the clipboard contents and sends the remainder to Safe Browsing for analysis. If the verdict comes back dangerous, Chrome puts a warning in front of the person before the command can be pasted and run. Google says the system is tuned for high precision so that developers who routinely copy terminal commands are not buried in false alarms.

Tuning matters because a developer copying an install line is behaviorally identical to a victim copying a fake one, and the browser has to tell them apart from the text alone. BleepingComputer’s report on the announcement describes the same flow: suspicious script, personal data stripped, Safe Browsing verdict, then a warning only if the result is dangerous.

What remains unproven

Google has not published detection rates, false-positive rates or the thresholds that separate a suspicious site from a dangerous one, and no outside tester has yet measured how well the clipboard check catches commands that attackers obfuscate. Attackers adapt quickly to any single control, and a warning that appears only for opted-in users leaves the default population untouched. What Google has established is the design: a model that reads pages, a clipboard check that fires before execution, and a middle tier of caution.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

The day in tech, in one email: Get TLDR →

We may earn a commission if you sign up.