Google DeepMind and Singapore fine-tune an on-device Gemma model to flag scam chats
Google DeepMind and GovTech Singapore announced a privacy-preserving, on-device scam-detection model fine-tuned on local scam conversations. The model is to run entirely on users' devices with consent, so neither Google nor the government can read users' messages through it.
Photo: Dietmar Rabich / Wikimedia Commons, CC BY-SA 4.0. Google DeepMind logo via Wikimedia Commons (public domain).
Google DeepMind said on October 14 that it has been working with the Government Technology Agency of Singapore (GovTech Singapore) under Singapore's National AI Partnership to fine-tune a privacy-preserving, on-device scam-detection model. The announcement was presented in connection with Singapore International Cyber Week 2026.
DeepMind said Google's lightweight Gemma model was fine-tuned on local scam conversations ranging from impersonations of public institutions, banks and local figures to regional schemes, so it can distinguish benign chatter from manipulation when deployed in citizen apps. Unlike cloud scanning of messages, the model is designed to process conversation data on-device and trigger a real-time warning advising users to pause before sharing sensitive information or funds.
The system uses a two-stage pipeline: a lightweight classifier screens inbound messages using minimal power, then the fine-tuned Gemma model analyzes a larger conversation snapshot to confirm a threat or discard a false alarm. DeepMind said roll-out within applications will be carefully phased with the Singapore government, and that when deployed the model will only access on-device conversations with the user's consent. Because it runs entirely on-device, "neither the Singapore Government nor Google will be able to access users' conversations through its deployment," the post said.
DeepMind said fine-tuning took place in a GovTech-managed GCP "sovereign sandbox," training excluded private or personally identifiable information and used anonymised datasets from scam reports and intelligence feeds, and the model was stress-tested against synthetic dialogues and local corpora. Mark Chen, Director of Trust and Safety at GovTech Singapore, said the partnership brings advanced AI protection onto citizens' devices while keeping personal communications private. DeepMind cited Singapore figures that scams make up over half of reported crimes and more than S$4 billion in losses since 2019, with social-engineering scams accounting for up to 80% of losses.