Google has announced a significant expansion of Android’s AI-driven scam protection capabilities, bringing on-device Gemini models to new hardware partners and a broader set of languages and regions. The updates target both real-time call fraud and increasingly sophisticated conversational messaging scams.
Scam Detection for Calls Comes to Samsung
Android’s Scam Detection feature, which analyzes live call audio for speech patterns associated with fraud, has until now been limited to Google Pixel devices. Google is extending the feature to the Samsung Galaxy S26 series in the United States, marking the first expansion to a third-party manufacturer. Pixel availability already covers twelve countries including the US, UK, Australia, Canada, France, Germany, India, Ireland, Italy, Japan, Mexico, and Spain.
The feature runs entirely on-device using Gemini’s on-device model, meaning audio is neither stored locally after processing nor transmitted off the device. Scam Detection is disabled by default, applies only to calls not originating from a user’s contact list, and can be toggled in system settings.
Smarter Messaging Protections in More Languages
Google Messages’ Scam Detection capability has expanded to more than 20 countries with support for Arabic, English, French, German, Portuguese, and Spanish, among other languages. In the US, Canada, and the UK, the system is being upgraded to use the Gemini on-device model on flagship Android hardware, beginning with the Pixel 10 series and extending to the Samsung Galaxy S26.
The Gemini-backed analysis is specifically designed to catch threat types that resist simple pattern matching. These include job-offer scams and so-called pig-butchering schemes, where attackers cultivate a fabricated relationship over weeks or months before steering victims toward fraudulent investments. Because these attacks unfold gradually and lack the abrupt warning signs of conventional phishing, detecting them requires contextual reasoning across a conversation thread rather than single-message analysis.
Scale and Independent Assessment
Google states its multi-layered defenses currently intercept more than 10 billion suspected malicious calls and messages per month across Android. An evaluation by Counterpoint Research cited in the announcement concluded that Android provides the most comprehensive AI-powered scam protections of any mobile platform, though Google did not publish detailed methodology from that study.
Privacy Architecture
Both call and messaging protections share a common design principle: all inference runs on the device. No conversation content is stored after processing or shared with external servers. Google frames this as a deliberate trade-off to maintain user privacy while still delivering real-time warnings, a constraint that also influenced the choice of Gemini’s on-device model rather than a cloud-based inference pipeline.
The messaging updates are rolling out now, with call detection expansion to Samsung hardware expected alongside the Galaxy S26 launch.