Singapore’s Government Technology Agency and Google DeepMind are developing an artificial intelligence model that will detect potential scams directly on Android devices.
The model will analyse conversational patterns in real time and display a warning when it identifies signs of social-engineering tactics, allowing users to pause before disclosing credentials or transferring money.
Join Singapore Business Review community
Unlike conventional systems that scan information on cloud servers, the model will conduct its analysis locally on the device.
This approach is intended to preserve the privacy of personal communications and allow the technology to work with messaging services protected by end-to-end encryption.
Lightweight Gemma models will be fine-tuned using local scam narratives, including impersonations of public institutions, banks and prominent individuals, as well as emerging regional schemes.
The system will also undergo continuous training to distinguish ordinary conversations from the structured and high-pressure language commonly used by scammers.
Mark Chen, director of trust and safety at GovTech Singapore, said the approach would allow targeted local scams to be identified without exposing users’ private communications.
Further details on how users can install and use the model will be announced later.
Singapore is also joining Google’s Fairwind Programme as part of the wider collaboration announced during Singapore International Cyber Week 2026.
The programme will give cyber defenders access to Google’s Gemini 4 Argon model and CodeMender harness.
According to Google, the tools can identify and verify vulnerabilities and generate deployment-ready patches within an organisation’s secure cloud environment, reducing a process that can take weeks to minutes.
The initiatives build on Google’s existing cooperation with Singapore, including a memorandum with the Cyber Security Agency that was refreshed in 2025 and an AI Agents Sandbox designed to study the operation of computer-use agents in real-world environments.
