Google DeepMind and Singapore Government collaborate to build fine-tuned AI model for scam detection
Google DeepMind and Singapore's GovTech agency fine-tuned a lightweight Gemma model on local scam conversations so it can run on phones and warn users when a chat looks manipulative, without sending messages to the cloud. A two-stage pipeline keeps battery use low, and training excluded personal data.
What the article says
Scams account for more than half of reported crimes in Singapore and have reportedly cost over S$4 billion since 2019, according to the page. Fraudsters have moved from simple phishing to social engineering, building a convincing conversation before asking for money; such cases are said to account for up to 80 percent of losses. Google DeepMind and GovTech Singapore, under their National AI Partnership with the government, have fine-tuned a lightweight Gemma model for on-device scam detection, trained on local conversations including impersonation of public officials, banks and local figures, and trending regional schemes.
When the model spots manipulation, it shows a real-time warning urging the user to pause before sharing sensitive information or money. Two stages limit battery use: a lightweight classifier screens messages, then the Gemma model analyzes a larger snapshot of the conversation. Analysis runs on the device, so neither the government nor Google can access conversations, and it works only with the user's consent. Rollout will be phased; the page gives no accuracy figures or partner apps.
Read the official article AnnuIA profile
Summary written by AnnuIA in its own words from the official page of Google DeepMind. The content and rights belong to the publisher: please read the original article.