AI

Google DeepMind and GovTech Launch On-Device Scam Detection Model

Wednesday, October 14, 2026Read Original

Details

  • Google DeepMind and Singapore’s Government Technology Agency, GovTech, are developing an on-device model to detect scam conversations in real time.
  • The system uses Google’s lightweight, open Gemma model, fine-tuned on Singapore-specific scam conversations and regional tactics.
  • Training examples include impersonation of public institutions, banks and local figures, alongside emerging local scam methods.
  • A two-stage pipeline improves efficiency: a low-power classifier first screens incoming messages, while Gemma analyzes a broader conversation snapshot when suspicious content is detected.
  • If manipulative patterns are confirmed, the device displays a warning advising users to pause before sharing sensitive information or transferring money.
  • Processing takes place on the smartphone rather than external servers, helping preserve the privacy of conversations, including chats handled through encrypted messaging services.
  • The model is expected to be introduced gradually through applications developed with the Singapore Government; no public release date, supported applications or performance metrics were announced.

Impact

The project positions small, locally adapted models as practical tools for public-sector cybersecurity rather than only general-purpose assistants. On-device processing could reduce privacy concerns and cloud costs, while Singapore-specific fine-tuning may improve relevance against local impersonation scams. Its impact will depend on false-positive rates, battery demands, device support and whether government-backed applications can deploy the system at scale.

Rift Dispatch
Google DeepMind and GovTech Launch On-Device Scam Detection Model | RiftLab