Google is deploying its Gemini Nano artificial intelligence model directly within the Chrome desktop browser, introducing a new layer of on-device scam detection designed to protect users in real-time. Starting with the upcoming Chrome 137 release, this feature leverages local processing to analyze potentially malicious websites, initially targeting deceptive tech support scams.
Instead of relying solely on cloud-based analysis, Chrome will now leverage the local processing power of Gemini Nano to evaluate webpage content when specific triggers, such as the use of the keyboard lock API, are detected. The AI model extracts security signals, like the page’s likely intent, which are then used by Google’s Safe Browsing system to make a final determination. Users who have opted into Chrome’s Enhanced Protection mode will see a warning if a site is deemed likely malicious.
This on-device approach offers crucial advantages for user security and privacy. It allows Chrome to react instantly to potentially harmful sites, including short-lived ones designed to evade detection, without needing to send the full page content to Google servers for the initial assessment. Google emphasizes that the process is designed with performance in mind, using techniques like asynchronous processing and resource throttling to minimize impact on browsing activity.
How The On-Device AI Protection Works
When a user navigates to a page exhibiting characteristics typical of tech support scams, Chrome 137 will activate the built-in Gemini Nano model. This lightweight AI, previously introduced on Pixel phones, analyzes the content locally to identify potential threats. According to Google, these AI-generated security signals – not the raw page data – are then sent to Safe Browsing for users enrolled in Enhanced Protection.
If Safe Browsing, combining the on-device signals with its existing intelligence, flags the site as a likely scam, Chrome will display a prominent warning, preventing the user from proceeding without acknowledging the risk. While the primary protection layer benefits Enhanced Protection users, Google notes that Standard Protection users will also gain indirectly as newly discovered malicious sites are added to Safe Browsing’s universal blocklists. TechCrunch reports that Google is also rolling out separate AI-powered warnings for potentially spammy or malicious notifications on Chrome for Android.
Building on Previous AI Integration
This security feature builds upon Google’s earlier integration of Gemini Nano into Chrome, first announced at Google I/O 2024. That initial phase focused on enabling developers to leverage the on-device model, powered by advancements in WebGPU and WebAssembly, for building AI features directly into web applications. Google itself planned to use it for tools like Gmail’s ‘help me write’ feature, showcasing the versatility of having the model available locally within the browser environment.
Navigating Broader Chrome Challenges
The deployment of Gemini Nano for scam detection represents a practical application of the on-device AI framework introduced previously. This development occurs within a challenging context for Google and Chrome.
The company recently halted its controversial plan to phase out third-party cookies, citing industry pushback and ongoing regulatory discussions, particularly in the UK. This decision followed difficulties with its Privacy Sandbox alternatives, highlighting the complexities of reshaping web technologies.
Simultaneously, Chrome remains a central focus in the remedies phase of the U.S. Department of Justice’s antitrust case against Google’s search monopoly. The DOJ is pushing for a forced sale of the browser, arguing it’s used to unlawfully maintain search dominance, a demand Google fiercely resists, with company lawyers calling it “extreme” and “fundamentally flawed”. This high-stakes legal battle even saw OpenAI express potential interest in acquiring Chrome should divestiture be mandated.
Future Roadmap And AI Safety Efforts
Looking ahead, Google plans to extend Gemini Nano’s scam detection capabilities beyond tech support schemes to include threats like package tracking and unpaid toll scams in the future. An expansion to Chrome on Android is also planned for later in 2025. Furthermore, Google is actively researching ways to counter potential adversarial attacks, such as prompt injection, aimed at circumventing the AI detection.
This Chrome feature aligns with Google’s broader strategy of using AI for safety, complementing existing efforts in Search where AI systems reportedly help block hundreds of millions of scammy results daily and have significantly reduced specific threats like fake airline support scams.
