Online scams: how Google uses AI to track them on Chrome, Android and Search

Alarming pop-ups, fake customer service numbers, trapped notifications or fraudulent calls: online scams are evolving and becoming more professional. Faced with this growing threat, Google has announced the rollout of a new generation of protections based on artificial intelligence. From Search to Chrome and Android, detection tools are becoming smarter, faster and better integrated.

 This article in audio format

A constantly evolving threat

Online scamming is a scourge that keeps adapting. Between fake virus alerts, entirely fabricated customer support numbers, cloned sites or alarmist messages, techniques are multiplying and growing more complex. In 2024, global losses related to scams exceeded $1 trillion, according to the Global Anti-Scam Alliance.

Faced with these increasingly sophisticated attacks, Google has decided to step up the pace by betting on its most powerful weapon: artificial intelligence. More specifically, on models capable of running directly on users' devices, without connecting to servers.

Cleaner search results

Google Search is often the entry point for many scams. That's why the company has invested heavily in AI systems able to detect suspicious pages at large scale.

Thanks to these technologies, Google now says it blocks 20 times more fraudulent results than it did a few years agowith entire campaigns of malicious sites neutralized as soon as they appear. One example among many: fake airline customer service pages that redirected users to premium-rate bogus numbers have fallen by 80% in search results.

AI here allows Google to analyze massive amounts of text, detect networks of sites linked to each other, and identify linguistic patterns typical of scams.

Chrome: safer browsing thanks to Gemini Nano

On Chrome, security is also shifting into high gear with Gemini Nano, a language model (LLM) that runs directly on your computer. This tool analyzes in real time the pages you visit and detects suspicious behavior, even that which evades traditional systems.

This local operation represents a major turning point: it guarantees an instant response and preserves the user's privacy, since the data never leaves the device.

Among the threats targeted by Gemini Nano, Google notably cites the technical support scams, where sites simulate a crash or a viral alert to trick the user into calling a fraudulent number. Gemini Nano identifies subtle signals such as the use of certain APIs, alarming texts, or keyboard-locking tactics. To limit resource consumption, the model runs asynchronously, with a controlled GPU usage quota.

Chrome on Android: deceptive notifications under surveillance

Google has also strengthened its protection system against abusive notifications on mobile Chrome. Some malicious sites exploit this feature to spam the user with deceptive messages, prompting them to click on malicious links.

A local AI model now detects these notifications and displays an alert. The user can then choose to unsubscribe or view the blocked content, with the option to restore notifications if the alert proves unfounded.

Android: strengthened security with Android 16

With Android 16, Google goes even further with a series of advanced features inspired by Apple's Lockdown mode. This new enhanced protection mode automatically enables several mechanisms:

• Disabling JavaScript, 2G connections and app sideloading;
• Smart lock in case the device is stolen;
• Blocking access to one-time passwords (OTP) while the phone is locked;
• Enhanced app scanning via Play Protect, able to detect apps that change their icon or hide their presence.

Another major innovation: the automatic alert during phone calls. If a user shares their screen with a stranger and opens a banking app, Android emits a suspected scam alert, in collaboration with several banks (Monzo, NatWest, Revolut).

Messages and calls: smarter alerts

On the Messages app and the Phone app, Google has also integrated a real-time scam detection. Thanks to AI, suspicious messages (about tolls, unpaid fees, fake winnings, cryptocurrencies, etc.) are automatically flagged.

The algorithm relies on recognition of textual structures, the sending context, and behavioral signals. In addition, a tool called Key Verifier allows you to verify your correspondent's identity by comparing encryption keys, via QR code or a shared code.

A multilingual AI for a global threat

One of the major strengths of the language models deployed by Google is their ability to operate in multiple languages. When a phishing campaign is detected in English, the AI can automatically identify its variants in Hindi, Spanish, German, or French.

This approach enables Google to respond quickly on a global scale, without waiting for each local version of a scam to be reported.

The article “Online scams: how Google uses AI to track them on Chrome, Android and Search” was published on the site Abondance.