Google DeepMind and the Government Technology Agency of Singapore (GovTech) announced an on-device scam detection model on October 14, 2026, presented at Singapore International Cyber Week. Scams make up more than half of Singapore's reported crimes, with losses above S$4 billion since 2019, and social engineering now accounts for up to 80 percent of losses. Read the announcement.
How the detector works
A lightweight Gemma model was fine-tuned on local scam conversations, covering impersonations of public institutions, banks and local figures as well as regional schemes. Detection runs in a two-stage pipeline: an efficient classifier first screens inbound messages using minimal power, and a contextual evaluation with the fine-tuned Gemma model then reviews a larger snapshot of the conversation to confirm a threat or discard a false alarm. When manipulative patterns are detected, a real-time warning is triggered on the device advising the user to pause before sharing sensitive information or funds.
Why on-device
Server-side scanning cannot inspect conversations on end-to-end encrypted messaging platforms. Processing on the device keeps analysis of conversation patterns local, which the partners describe as faster detection of manipulative intent without exposing private communications.
Governance
- Fine-tuning took place in GovTech's managed Sovereign Sandbox on Google Cloud, keeping training data isolated.
- Training excluded private and personally identifiable information, using anonymised datasets from scam reports and intelligence feeds.
- The model was stress-tested against synthetic dialogues and local corpora covering diverse scam categories.
The announcement positions the project as a template for messaging platforms and operating system vendors, which can integrate such lightweight models locally. Related coverage of model safety practices is in our source verification guide.