Google DeepMind and Singapore develop scam detection AI that runs on phones

Google DeepMind announced on October 14 that it has developed a scam detection model with Singapore's GovTech agency. The system analyzes conversations on a user's device and is designed to warn about manipulation before money or sensitive information changes hands.

The partners adapted Google's Gemma model using examples of local scams, including impersonation of banks and public officials. An initial filter screens messages, then Gemma examines more conversation context when something looks suspicious. Google said this two-step approach helps limit battery use.

The project addresses fraud that develops through persuasive exchanges rather than a single suspicious message. Google said processing stays on the device, preventing Google and the government from accessing conversations through the system.

Deployment through applications will proceed in stages with Singapore's government. Access to conversations will require user consent under each application's terms, and the announcement gives no specific public launch date.