Scams now make up more than half of Singapore's reported crimes and have caused over S$4 billion in losses since 2019, according to Google DeepMind. The company notes that social engineering tactics—where attackers build convincing conversations before asking for money—now account for up to 80% of all losses. To address this, DeepMind worked with Singapore's Government Technology Agency under the country's National AI Partnership to build a scam detection model that runs locally on a user's device.
The model is a fine-tuned version of Google's lightweight Gemma model, trained on local scam conversations such as impersonations of banks, public institutions, and local figures. Because it runs on-device, it can analyze conversation patterns quickly and issue real-time warnings before users share sensitive information or transfer funds. To manage battery use, the system uses a two-stage pipeline: a low-power classifier first screens incoming messages, and then the fine-tuned Gemma model examines a larger conversation snapshot to confirm a threat or dismiss a false alarm.
DeepMind emphasizes that privacy was central to the design. Fine-tuning took place inside GovTech's managed cloud environment, training data excluded personally identifiable information, and the model was stress-tested against synthetic dialogues and local corpora. When deployed, the model will only access on-device conversations with user consent, and neither Google nor the Singapore government will be able to see those conversations. The project was presented at Singapore International Cyber Week 2026, and DeepMind says it is exploring similar deployments in other countries.