On October 20, 2026, Google announced new scam defenses for Chrome's Enhanced Protection feature. The update expands warnings for sites suspected of being dangerous and moves Gemini-based scam detection from an on-device model to a server-side model. It also adds protection against "ClickFix," a technique that displays fake repair instructions to trick users into running malicious commands. Chrome can now warn users at the moment a suspicious command is copied. Attackers can bypass the browser's download protections by using built-in OS tools to download malware, so Google is extending detection from actions inside the browser to commands that may be run afterward. At the same time, moving AI analysis to the server improves detection but raises an important question: what data is sent to Google?
A new site warning that distinguishes "Suspicious" from "Dangerous"
Google has split website warnings in Chrome's Enhanced Protection into two levels: "Suspicious" and "Dangerous." In addition to the existing warning for dangerous sites, Chrome now also alerts users to sites that may harm them even when there isn't enough evidence to conclude they are malicious. Google has long used a similar distinction when assessing the safety of downloaded files, and it has now extended that approach to website warnings.
The new "Suspicious" warning appears when Chrome detects deceptive patterns commonly seen on known malicious sites. It targets sites that could be used for scams, even though confidence that they are malicious is relatively low. Sites judged malicious with high confidence and posing a high risk of harm to users continue to get the full-screen red "Dangerous" warning. The two warnings differ in format, icon, color, and wording, so users can tell how serious the risk is.
The goal is to give users appropriate caution even before a site is confirmed as malicious. Showing the same strong warning for every questionable site would make it hard to tell them apart from sites that are definitely dangerous, but varying the strength of the warning lets Chrome cover a broader range of sites. Google says that adjusting how warnings are displayed allows it to widen the set of sites flagged without compromising user safety. However, the October 20 announcement did not disclose a false-positive rate for the new "Suspicious" determination, nor specific figures on how much the number of warned sites has increased.
Gemini scam detection moves from the device to the server
The biggest change in this update is how Gemini is used to detect scam sites. In 2025, Google introduced an on-device Gemini Nano model to Enhanced Protection to detect tech support scams. The new approach instead uses a more capable server-side Gemini model. This extends AI-based detection beyond supported devices to all users who have Enhanced Protection turned on, and it adds coverage for scams that pose as fake antivirus software or browser updates.
The 2025 approach and the new one differ not only in where the AI analyzes the page but also in what data is sent to Google. Comparing Google's May 8, 2025 technical explanation with the October 20, 2026 announcement gives the following:
| Item | May 2025 approach | October 2026 approach |
|---|---|---|
| Where the page is analyzed | Gemini Nano on supported devices | Server-side Gemini model |
| What triggers analysis | Behavior typical of support scams, such as APIs that lock keyboard input | Signs of a scam detected with high confidence |
| What is sent to Google | Security-relevant features extracted by the AI, such as the page's intent | Text from the page being analyzed |
| Final risk determination | Safe Browsing combines the AI-extracted features with other information | Safe Browsing combines Gemini's analysis with signals such as domain reputation |
Under the 2025 approach, when Chrome detected behavior typical of tech support scams, such as use of an API that restricts keyboard input, the on-device Gemini Nano analyzed the page. The AI extracted information indicating things like the page's intent and sent the result to Google Safe Browsing to assess the risk. Under the new approach, when signs of a scam are detected with high confidence, the page text is sent to the server-side Gemini for analysis. In other words, even the 2025 approach sent information needed for safety determinations to Google, so this is not the first time external communication has occurred.
The new Gemini model analyzes which company or brand a page is impersonating, what it is trying to get the user to do, and how likely it is to be a scam. Safe Browsing then combines those results with security signals such as domain reputation to make the final risk determination. Rather than showing a warning based on the AI's judgment alone, the system builds the AI's ability to understand page content into existing security assessments.
Regarding privacy with server-side analysis, Google says the design minimizes the data sent. Page text is analyzed only when signs of a scam are detected with high confidence. Google also says that page content sent off the device is processed only temporarily, is not logged or stored, and is not used to train AI models. However, this explanation applies to the content sent for the Gemini page analysis introduced in this update.
Enhanced Protection as a whole shares a broader range of data with Google. According to Chrome's official help page, when Enhanced Protection is on, Safe Browsing receives the URLs you visit as well as page content, extension activity, and some system information. If you sign in to your Google Account and also turn on account-level Enhanced Protection, data is saved to the account and protection is strengthened in Google services such as Gmail. The "no storage" policy described for the Gemini page analysis therefore cannot be applied to everything sent under Enhanced Protection.
ClickFix defense detects malicious commands when they are copied
Another important change addresses "ClickFix," a social engineering attack that has spread in recent years. ClickFix displays a fake error or security warning on a website and, supposedly to fix the problem, tricks the user into copying and running a command. The command shown on the page may differ from what is actually copied to the clipboard. These attacks often abuse legitimate tools that ship with the OS, such as Windows PowerShell and cmd.exe.
What makes these attacks troublesome is that the malware is downloaded and run outside the browser. With an ordinary malicious file download, Chrome and Safe Browsing have a chance to check the file's safety. But when the user runs PowerShell or a similar tool themselves and the command fetches the malware, the download proceeds through OS tools and can evade the browser's download protections. By abusing legitimate system tools, attackers try to blend suspicious activity into ordinary administrative work.
To counter this, Google has added to Enhanced Protection a mechanism that detects the copying of suspicious commands. When a user copies a script that could be used in ClickFix, Chrome strips personally identifiable information from the clipboard contents and sends it to Safe Browsing for analysis. If the command is judged dangerous, Chrome shows a warning. This lets Chrome alert users to the risk before they paste the command into a terminal and run it.
This does not mean Chrome constantly monitors all clipboard contents and sends them to Google. What Google described is a mechanism that analyzes content when a suspicious command is copied, and it says the design aims for few false positives to limit the impact on legitimate development work. However, Google did not disclose how personal information is removed, how long the sent copy contents are retained, or the actual false-positive rate. Whether the no-storage policy described for Gemini page analysis also applies to data sent for ClickFix defense cannot be confirmed from this announcement.
This defense also detects the copying of suspicious commands and warns according to risk; it is not a feature that necessarily blocks execution on the OS. Nor does Google describe any mechanism for remediating a malware infection caused by a command that has already been run. The change gives Chrome an opportunity to intervene at the copy step, just before a command is run outside the browser, an area its protections have had difficulty reaching.
The basis for Google's "10x safer" claim, and choosing a protection level
In this announcement, Google claims that users of Chrome's Enhanced Protection are "10x safer" than Standard Protection users against scams such as phishing, social engineering, and malicious notifications. The way that figure was calculated deserves attention, though. According to the announcement's footnote, it is based on a comparison of the average number of safety warnings shown to Enhanced Protection users and Standard Protection users. It is not a direct comparison of actual scam losses or malware infection rates.
Google also has not disclosed details such as the period compared, the number of warnings in each group, or how user conditions were matched. This figure alone therefore cannot support the conclusion that enabling Enhanced Protection reduces the chance of falling victim to a scam by 90%. Nor does the metric reveal how much harm each of the newly introduced features, the suspicious-site warnings, Gemini analysis, and ClickFix defense, actually prevents. An increase in warnings shown must be evaluated separately from a decrease in actual harm.
In Chrome, Standard Protection is the default, and the new features described here apply to users who have turned on Enhanced Protection. In desktop Chrome, you can change the Safe Browsing protection level by going to Settings, then Privacy and security, then Security. Standard Protection still provides warnings about known dangerous sites and downloads, while Enhanced Protection adds extra checks for unknown threats and AI-based analysis. However, Google's October 20 announcement did not give a detailed rollout schedule for each OS and region, or the minimum Chrome version required.
With this update, Chrome's scam defenses go beyond detecting known dangerous sites and files: they analyze page content with AI and can warn users even before they run a malicious command. At the same time, because Enhanced Protection sends some of the information needed for those determinations to Google, users need to understand detection capability and data-sharing conditions together. If false-positive rates, the harm prevented by detection, and the storage conditions for data sent by the ClickFix defense are disclosed in the future, individuals and organizations will be able to evaluate the effectiveness of the new protections more concretely.