Google moves Chrome’s AI scam checks to servers and adds command warnings

Chrome users with Enhanced Protection enabled can now get warnings about suspicious websites and copied commands that may install malware. In an October 20 announcement, Google detailed its updated scam defenses, including a shift to server-side Gemini analysis that brings AI checks beyond the supported devices covered by its earlier on-device system.

Google describes the package as features introduced during 2026, rather than a single release arriving today. They address different points in a scam: judging a page’s intent, warning when evidence is suspicious but not conclusive, and detecting commands copied from a fake repair prompt.

Gemini moves off the device

Last year, Chrome added on-device Gemini detection for tech-support scams on supported devices. Google says the move to a more powerful server-side model extends that protection to all users who turn on Enhanced Protection. It also broadens the target to threats such as fake antivirus notices and fraudulent browser-update prompts.

The model examines page content for the brand being represented, the page’s apparent purpose and the likelihood of a scam. Gemini does not make the final decision alone: Safe Browsing combines its analysis with the domain’s reputation and other security signals before returning a verdict.

That change sends some page text off the device. Google says this happens only when a site already shows high-confidence scam indicators. It says the content is processed temporarily and is never logged, retained or used to train models. Those are the company’s stated privacy safeguards for the server-based checks.

Suspicious is not the same as dangerous

Chrome’s new warning system separates lower-confidence suspicions from high-confidence danger. A suspicious-site warning appears when a page shows deceptive patterns associated with known malicious sites. Google says these warnings differ in format, icons, color and wording, giving users a distinct signal about possible harm rather than treating every flagged page identically.

Sites judged dangerous retain Chrome’s familiar full-page red warning. Google says distinguishing the two levels lets it show more protective warnings while making clear which verdicts carry greater certainty and risk. The design follows the suspicious-versus-dangerous distinction Chrome already uses for download warnings.

Catching a scam before the command runs

The clipboard defense targets ClickFix, a trick that asks people to copy and run a command to repair a fake problem. Google says the copied command often differs from the one displayed on the page. Following the instructions can hand execution to the computer’s own tools instead of the browser.

Attackers repurpose legitimate, pre-installed programs such as PowerShell or cmd.exe to download and set up malware. Google says shifting the download into a terminal—a program for running typed commands—sidesteps Chrome and Safe Browsing’s existing download protections. The malicious action takes a route those browser checks were not covering.

Chrome now detects suspicious commands copied by Enhanced Protection users. A potential ClickFix payload is stripped of personally identifiable information and sent to Safe Browsing for analysis. If the result is dangerous, Chrome displays a warning. Google says the system is designed for high precision to avoid disrupting legitimate developer work.