Google DeepMind and Singapore launch on-device scam detection tool

The solution uses a lightweight version of Google’s Gemma model, fine-tuned on local scam conversations to distinguish between benign chatter and manipulative intent. Unlike traditional cloud-based scanning, this model runs directly on smartphones, preserving user privacy by keeping conversation data on the device. It employs a two-stage pipeline: a low-power classifier screens messages, followed by a deeper contextual analysis by the Gemma model to confirm threats and trigger real-time warnings.

The project was presented at Singapore International Cyber Week (SICW 2026) and emphasizes strict governance, including training within a sovereign sandbox and excluding personally identifiable information. The initiative aims to provide a template for other messaging platforms and operating system vendors to integrate local safety features without compromising end-to-end encryption.