Google plans to introduce a new security feature in Chrome 137 that uses on-device AI to detect tech support scams in real time.
The new system leverages the Gemini Nano large language model (LLM) to bolster Safe Browsing's ability to catch rapidly evolving scam tactics.
While Google Safe Browsing has long served as a frontline defense against malicious sites, the new integration of LLM-powered signal generation offers a dynamic response to threats that appear and vanish within minutes — too fast for traditional crawler-based detection to catch.
The development comes amid a sharp rise in tech support scams, where attackers mimic security alerts to trick users into paying for fake services or surrendering control of their devices. These scams frequently use aggressive tactics, such as full-screen overlays, keyboard and mouse input suppression, and alarming pop-ups, to manufacture a sense of urgency. According to Google, the average lifespan of a malicious scam page is under 10 minutes, underscoring the need for near-instantaneous detection.
The new feature in Chrome 137 uses the on-device Gemini Nano LLM to evaluate suspicious web pages directly on users' devices. When certain behaviors, like abuse of the keyboard lock API, are detected, the browser invokes the LLM to analyze page content and infer intent. The results are summarized into security signals and shared with Safe Browsing servers for a final verdict. If a threat is confirmed, users are immediately warned via a full-page interstitial alert.
Google emphasizes that performance and privacy are safeguarded throughout the process. The LLM runs locally and only when specific conditions are met. Its operations are throttled and constrained to limit GPU resource use and prevent interference with the browsing experience. Moreover, LLM-generated signals are transmitted only for users who have opted into Chrome's Enhanced Protection mode, ensuring that privacy-conscious users retain control over their data.
Looking ahead, Google plans to expand the use of on-device AI to counter other fast-growing scam categories, such as fake package tracking alerts and unpaid toll scams. Chrome on Android is also slated to receive the LLM-based protections later this year.
To stay safe, users are encouraged to enable Enhanced Protection in Chrome's security settings, which not only activates this new AI-powered layer but also ensures more frequent security updates and proactive threat identification.
Leave a Reply