On-device scam detection model with GovTech Singapore
Google DeepMind and GovTech Singapore fine-tuned a lightweight Gemma model on local scam conversations to detect scams on-device, with no chat data leaving the phone.
The model is a Gemma model fine-tuned on anonymised local scam data, such as impersonation of public institutions and banks. It runs in two stages. A low-power classifier screens inbound messages, then the fine-tuned Gemma model reviews a larger snapshot of the conversation to confirm a threat or discard a false alarm. Rollout inside applications will be phased and has not started.
- Date
- Wednesday 14 October 2026
- Lab
- Google DeepMind
- Kind
- model
- Access
- research preview
The post does not name the Gemma size or version, give accuracy figures, or say which apps will use the model. It was presented at Singapore International Cyber Week 2026.
Sources
This record was checked and corrected against its sources on 14 October 2026. How we check