On Thursday, Google announced it will deploy new AI (AI) power measurements to combat fraud via Chrome, Search and Android.
Tech Giant said it will begin using Gemini Nano, its on-device leading language model (LLM) to improve secure browsing of Chrome 137 on desktops.
“The on-device approach can provide immediate insight into risky websites and provide protection against fraud that you’ve never seen before. GeminiNano’s LLM is perfect for this use as it has the ability to distill the nature of various complex websites and helps you adapt quickly to new fraud tactics,” the company says.
Google said it is already using this AI-driven approach to tackle remote technical support scams. This is often a request that users split their personal or financial information under the pretext of non-existent computer problems.
This works by using LLM to evaluate web pages for potential signals that are symbolic of technical support fraud, such as the use of the keyboard lock API. The security signal is then extracted and passed to secure browsing to determine if the page is likely to be fraudulent.
“In addition to ensuring that LLM is conservatively triggered and executed locally on the device, it carefully manages resource consumption by taking into account the number of tokens used, running the process asynchronously to avoid interrupting browser activity, and implementing throttling and quota enforcement mechanisms to limit GPU usage, avoiding implementation of throttling and quota enforcement mechanisms.
 |
| An overview of how devices’ LLM support works in fraud mitigation |
Google said it plans to expand this feature to detect other types of fraud, including those related to package tracking and unpaid tolls. This feature is expected to be rolled out to Chrome on Android later this year.
As part of the announcement, Google has bolstered its AI-powered fraud detection system to eliminate 20x deceptive pages, block such pages from search results, reduce schemes that impersonate airline customer service providers over 80%, and reduce schemes that mimic official resources such as visas and government services in 2024.
Finally, Google said it is launching a new warning feature for Chrome on Android. This uses machine learning models on the device to warn users of unnecessary notifications sent by malicious sites that aim to download suspicious software or provide sensitive data.
“This new feature allows you to use machine learning on your device to further control the information you see on your device, in order to detect and alert you about potentially deceptive or spam notifications.”
“When a notification is flagged in Chrome, the site’s name sends a notification, and you’ll receive a message warning that the notification’s content is incorrectly fraudulent or spam, as well as an option to unsubscribe from the site or view flagged content.”
This feature has been just over two months after Google deployed an AI-powered fraud detection feature in its messaging app for Android. Last year, the company announced a similar method for flagging fraud calls.
Also, updates will arrive as Google appears to be preparing advanced Android 16 protection. This reflects Apple’s approach by turning off JavaScript, disabling 2G connections, disabling 2G connections by default, including theft detection lock, offline device lock, Android Safe Browsing, Spam protection, and more.
Google has also been found working on a feature to detect fraud sharing victims to open bank apps while on calls, Android Authority reported earlier this week.
