Android rolls out Gemini-powered scam protection in Google Messages

Android will begin rolling out an "enhanced protection" mode in Google Messages over the coming weeks, using cloud-based Gemini models to analyze the intent of unknown senders in real time. The feature extends the existing fraud-detection system and activates only with explicit user consent.

The system runs two engines in parallel. Gemini in the cloud examines conversational context, while Gemini Nano on the device scans suspicious links before the user opens them. Google says scammers often mask their true aims behind "slow-cooking" conversations that build gradually, and cloud analysis can catch patterns that on-device protection alone might miss. On privacy, Google says data is encrypted during processing and handled in a "dedicated secure space" the company itself cannot access, while the local model processes network signals without sending data off the device.

Beyond messaging, the announcement adds two more safeguards: blocking apps from initiating call forwarding and alerting the user when an attempt occurs, and automatically hiding sensitive information during screen sharing in Chrome and Android WebView. These changes close known vectors that attackers exploit to hijack accounts or steal credentials.

Fraud detection in chat notifications, first launched on Pixel last year, is now expanding to devices from Honor, Motorola, Oppo, OnePlus, Realme, Tecno, Vivo, Xiaomi and additional manufacturers. Google says it plans to bring the capability to hundreds of millions more users later this year.

The move marks a significant shift from signature- and pattern-based security toward real-time conversational context understanding, without surrendering local processing for the most sensitive components. The combination of cloud and device, with end-to-end encryption in an isolated environment, could become a reference model for other messaging apps facing the same problem.