Google DeepMind and GovTech unveil on-device scam detection model

Google DeepMind and Singapore's GovTech announced an on-device scam detection model built on a fine-tuned Gemma model that warns users about manipulative conversations without sending messages to the cloud. Scams have cost Singapore over S$4 billion since 2019, with social engineering driving up to 80% of losses.

• Model is a fine-tuned Gemma trained on local Singaporean scam conversations
• Two-stage pipeline pairs a lightweight classifier with contextual Gemma checks
• Training excluded personal data and ran in GovTech's isolated cloud environment
• Scam losses in Singapore have exceeded S$4 billion since 2019