Google DeepMind and the Government Technology Agency of Singapore (GovTech Singapore) said on October 14 that they have fine-tuned a Gemma model to detect scams directly on a user’s device.
How it works
The partners trained a lightweight Gemma model on local scam conversations, including impersonations of public institutions, banks and local figures. Because it runs on-device, message data is not sent to a server, and the companies say neither the Singapore Government nor Google can read users’ conversations through it. It can access chats only with the user’s consent, under the terms of the app that integrates it.
DeepMind cites Singaporean figures that scams make up over half of reported crimes and have caused more than S$4 billion in losses since 2019, with conversational social engineering accounting for up to 80% of losses. The rollout will be phased and developed with the government; no launch date was given.
Why it matters
It is a concrete case of on-device AI used for consumer protection where privacy is the main constraint.