Google DeepMind and Singapore partner on on-device scam detection model

On 14 October 2026, Google DeepMind announced a partnership with Singapore's Government Technology Agency (GovTech) to fine-tune a lightweight Gemma model into an on-device scam detection model that runs on ordinary smartphones. The model analyses conversations locally on the device and shows a real-time warning when it detects manipulative language, using a two-stage pipeline of a lightweight classifier for initial screening plus Gemma for contextual assessment to keep power consumption down. Fine-tuning was completed in a GovTech-managed GCP environment, and the training data excluded private or identifiable personal information, using only anonymised datasets. The current status is that the model has been released; reports do not mention deployment scope or launch timing.

Written by AI from the reports · updated 12 h ago