Google DeepMind and GovTech Singapore have announced a collaboration to build a fine-tuned AI model aimed at detecting scams. According to the description, the approach relies on lightweight, on-device Gemma models to intercept conversational scams within private chats. A stated design goal is to keep processing local, with no chat data sent to the cloud.
Why it matters
Running scam detection on-device with zero cloud data egress is positioned to address privacy concerns tied to scanning private messages. By keeping data local, the system aims to flag conversational scams without transferring users’ chat content to external servers.
Who should care
Government agencies, product teams working on messaging safety, and developers interested in on-device model deployment may find the collaboration relevant, particularly those focused on fine-tuning compact models for privacy-sensitive tasks.