Google DeepMind and GovTech Singapore build on-device Gemma scam detection

Google DeepMind and GovTech Singapore are building an on-device scam detector on Gemma, Minister for Digital Development and Information Josephine Teo announced Wednesday at Singapore International Cyber Week. The pitch: catch manipulation in private chats without sending a message to a server. "If our devices can recognise signs of manipulation early, we can help potential victims avoid financial or credential loss, while keeping their personal messages on their devices," she said.

The numbers, per DeepMind: scams make up over half of Singapore's reported crimes and account for more than S$4 billion in losses since 2019. Conversational social engineering now drives up to 80% of losses.

The design is a two-stage pipeline. A lightweight classifier screens inbound messages on minimal power. If something looks off, a Gemma model fine-tuned on local scam conversations reads a larger slice of the chat and either confirms the threat or discards the false alarm. A confirmed threat triggers a real-time warning telling the user to pause. DeepMind says fine-tuning ran in GovTech's managed GCP environment, with training that excluded private or personally identifiable information.

Context: GovTech and the Singapore Police Force already run a separate system that assesses more than 600,000 websites a day, according to CNA.

My take: this is the right architecture for end-to-end encrypted apps, because a server can't scan what it can't read. But the announcement gives no rollout date, no named messaging platforms and no minimum phone specs. A ministerial announcement is not a rollout plan. The architecture gets my vote; the installation calendar remains blank.