Google DeepMind and Singapore’s Government Technology Agency (GovTech) have released an on‑device scam‑detection model for smartphones. The tool scans incoming messages locally, flags suspicious conversational patterns and shows a real‑time warning. Because the analysis stays on the phone, it works even with end‑to‑end encrypted apps and does not send private data to the cloud.
How the model works
The system uses a two‑stage pipeline. Stage 1 is a very light classifier that screens each inbound message using minimal battery power. If the message looks risky, Stage 2 runs a fine‑tuned version of DeepMind’s lightweight Gemma model on a larger snippet of the conversation to confirm the threat. The model was trained on anonymised local scam reports, without any personal or identifiable information.
Why it matters for everyday users
Scams account for more than S$4 billion in losses in Singapore since 2019 and now represent up to 80 % of all financial‑crime losses. By detecting manipulative messages before users act, the tool aims to reduce those losses while preserving the privacy of encrypted chats.
Why it matters
For a typical Singapore user, the on‑device detector can warn them before they send money or personal details to a fraudster, potentially saving them from costly scams. Because the analysis never leaves the phone, users keep the privacy guarantees of encrypted messaging. The real impact on overall scam loss rates remains to be seen as the tool rolls out.
