Google DeepMind and GovTech Singapore fine-tune Gemma model for on-device scam detection

Google DeepMind and the Government Technology Agency of Singapore (GovTech Singapore) collaborated to develop a privacy-preserving, on-device scam detection model using a fine-tuned lightweight Gemma model. This model addresses the issue that scams, including investment fraud and fake job offers, constitute over half of Singapore’s reported crimes, accounting for more than S$4 billion in losses since 2019, with complex social engineering now causing up to 80% of losses. The on-device system processes conversation data locally, ensuring neither the Singapore Government nor Google can access user conversations, and uses a two-stage pipeline—a lightweight classifier (Stage 1) followed by in-context analysis with the fine-tuned Gemma model (Stage 2)—to provide real-time warnings to citizens.