Google DeepMind and Singapore GovTech ship on-device Gemma scam detector on October 14, conversations never leave device

Google DeepMind and Singapore Government Technology Agency (GovTech) jointly announced on October 14 a partnership to build an on-device scam detection model for Android that analyses conversational patterns in real time and shows a warning when it identifies signs of social-engineering tactics so users can pause before disclosing credentials or transferring money. The model runs entirely on the device, meaning conversations never leave the phone, which makes the design fit for end-to-end encrypted messaging apps. The two-stage pipeline runs a lightweight classifier first to screen inbound messages at very low power cost, then a fine-tuned Gemma model reviews a larger snapshot of any flagged conversation; the Gemma model is fine-tuned on Singaporean scam narratives including impersonations of public institutions, banks and prominent individuals and emerging regional schemes. Mark Chen, director of trust and safety at GovTech Singapore, said the approach allows targeted local scams to be identified without exposing users' private communications. Singapore is also joining Google's Fairwind cyber defender program to access Gemini 4 Argon and CodeMender inside a secure cloud enclave to identify and verify vulnerabilities and generate deployment-ready patches. The deployment timetable and user onboarding details will be published later.