What happened
On October 14, 2026, Google DeepMind revealed a privacy-focused scam detector that runs locally on hardware. It was built by refining its compact Gemma model using local scam chats. GovTech Singapore collaborated on the project under the National AI Partnership, and training left out any private or personally identifiable details. Deployment inside apps will roll out gradually, developed alongside the Singapore Government. The model reads on-device conversations only when users agree, subject to each app's terms of use.
Why it matters
For people using apps that integrate the model, conversation content is handled directly on their handset, and the model issues an instant alert on screen urging them to stop and proceed carefully.
Read the original at Google DeepMind