Google DeepMind says it built an on-device scam detector for Singapore

Google DeepMind and Singapore’s Government Technology Agency (GovTech Singapore) say they developed a lightweight Gemma model, fine-tuned on local scam conversations, to identify manipulation in chats directly on the device. Google DeepMind says scams account for more than half of Singapore’s reported crimes and have caused over S$4 billion in losses since 2019; social-engineering scams account for up to 80% of losses.

The model will be rolled out gradually in applications in partnership with the government. When deployed, it will access on-device conversations only with the user’s consent, and neither the government nor Google will be able to access them through the deployment. The system combines a low-power classifier with broader analysis using Gemma, which can issue real-time warnings. Training took place in a GovTech-managed environment and excluded private or personally identifiable information, Google DeepMind says.