Google DeepMind, Singapore Build On-Device Scam AI

Google DeepMind and Singapore’s GovTech have developed an on-device AI model intended to detect scams in conversations, keeping message analysis on the user’s phone rather than sending it to external servers, the companies said on October 14. ^[1] CNA reported the collaboration and said the model is intended for Android devices. ^[2]

The lightweight Gemma model was fine-tuned on local scam conversations, including impersonations of public institutions and banks. ^[1] Google describes a two-stage process: a fast classifier screens messages, then the fine-tuned model analyzes a wider conversation snapshot when something appears suspicious. ^[1] The companies say an app using the model could warn users to pause before sharing money or sensitive information. ^[1]

The system has not been presented as a generally available service: Google says deployment within applications will be phased with the Singapore Government, while CNA reported that installation and usage details are still to come. ^[1][2] The announcements provide no detection-accuracy results, so the model’s effectiveness in real-world use remains unestablished. ^[1]