What we know

Google DeepMind and Singapore’s Government Technology Agency (GovTech) have developed an AI tool that detects conversational scams in private chats using lightweight on-device Gemma models. This technology analyzes messages locally on users’ devices, preventing any data from being sent to the cloud. The collaboration aims to enhance scam detection while maintaining strong privacy protections by eliminating cloud data egress.

Why it matters

Conversational scams through private messaging apps pose growing risks to users worldwide, including in Singapore. By running AI models directly on devices, this tool can identify scam attempts without compromising user privacy or relying on cloud processing.

What is still unknown

Details about the tool’s accuracy, the range of supported languages, and plans for wider deployment have not been disclosed. No independent evaluations of its effectiveness or privacy safeguards have been published.