The Government Technology Agency of Singapore and Google DeepMind are creating an AI model that will identify possible scams directly on Android devices.
The model will examine dialogue patterns instantly and show an alert when it detects indicators of social-engineering strategies, enabling users to halt before sharing credentials or sending funds.
In contrast to traditional systems that check data on cloud servers, the model will perform its analysis directly on the device.
This method aims to safeguard the confidentiality of private messages and enable the technology to function with messaging platforms secured by end-to-end encryption.
Lightweight Gemma models will be adjusted using local scam stories, incorporating impersonations of public entities, banks, and notable figures, along with new regional schemes.
The system will additionally receive ongoing training to differentiate regular conversations from the formal and intense language typically employed by scammers.
Mark Chen, trust and safety director at GovTech Singapore, stated that the strategy would enable the identification of specific local scams without compromising users’ private messages.
Also Read: LG Electronics Emerges as Apple's Home AI Ally
More information on how users can set up and utilize the model will be revealed at a later time.
Singapore is also participating in Google’s Fairwind Program as part of the broader partnership revealed during Singapore International Cyber Week 2026.
Also Read: FADU Unveils World's First File-Semantic SSD to Power AI Workloads
The program will provide cyber defenders with access to Google’s Gemini 4 Argon model and the CodeMender tool.
Google states that the tools can detect and validate vulnerabilities and create deployment-ready patches inside an organization’s secure cloud environment, shortening a process that typically takes weeks into mere minutes.
Also Read: EAM Jaishankar Urges Urgent Multilateral Reform, Cites India-UAE Ties
The initiatives leverage Google’s ongoing partnership with Singapore, which includes a renewed memorandum with the Cyber Security Agency from 2025 and an AI Agents Sandbox aimed at examining how computer-use agents function in real-life settings.