Google DeepMind and GovTech Singapore have worked together to build a fine-tuned AI model for scam detection. Google DeepMind describes it as using lightweight Gemma models that run on the device itself to intercept conversational scams in private chats.
The design is described as having zero cloud data egress, meaning private chat content is not sent out to the cloud for analysis. The source summary gives no accuracy figures, launch date, or list of supported apps and phones. It also does not say whether the tool would reach Hong Kong users.
What on-device scam detection means for families
- The focus is conversational scams in private chats, so the model is aimed at back-and-forth messages rather than public content.
- Because the Gemma models are lightweight and run on-device with zero cloud data egress, privacy-minded families have a concrete design claim to weigh.
- GovTech Singapore's involvement makes this a government-linked effort, but with no rollout date stated, families cannot yet assume it will be available on their phones.