Google DeepMind and the Singapore Government are collaborating to develop an on-device artificial intelligence (AI) model designed to detect and alert Android users to potential scams in real time. The initiative aims to protect users from evolving scam tactics while safeguarding personal privacy by processing data entirely on users’ devices.
The partnership with the Government Technology Agency of Singapore (GovTech) was announced on October 14 during the Singapore International Cyber Week. The AI system is engineered to analyze conversations within popular messaging applications like WhatsApp and Telegram, identifying local scam patterns, including attempts involving impersonation of government officials and financial institutions.
Mark Chen, director of trust and safety at GovTech, emphasized the balance between security and privacy, stating that the technology allows for immediate identification of highly targeted scams without exposing personal communications. Google DeepMind representatives added that the model operates solely on-device and only accesses conversations with the user’s consent, ensuring no data is transmitted to external servers or accessible by either Google or the Singapore Government.
The AI model has been developed using anonymized data from scam reports and intelligence feeds, and it has undergone extensive testing using synthetic conversations designed to mimic local speech patterns. Google highlighted that the system is continuously updated to respond to emerging scam methods, improving its accuracy in distinguishing legitimate exchanges from manipulative attempts.
When suspicious activity is detected, the model triggers a warning on the user's device, advising caution against sharing sensitive information or transferring funds. This approach, they said, represents a first-of-its-kind in real-time scam detection embedded within consumer devices, tailored specifically for Singapore’s scam landscape.
Singapore has faced significant losses to scams in recent years, with over S$4 billion lost since 2019. In the first half of 2026 alone, reported losses amounted to S$410.6 million. Common scams include those related to e-commerce, phishing, fraudulent investments, job offers, and impersonation of government officials.
In addition to the scam detection model, Google announced it would provide the Singapore Government access to its advanced AI cybersecurity tools. These tools are intended to help cyberdefense teams quickly identify and patch vulnerabilities, reducing risks of disruption to essential systems.
The collaboration aligns with Singapore’s strategic push for AI-driven cybersecurity measures amid increasing concerns about cyber threats and incidents involving AI agents breaching secure environments elsewhere in the world. Notable breaches, such as a rogue AI bot accessing an Australian healthcare database earlier this year, have underscored the importance of proactive measures in the cybersecurity domain.
A release date for the AI scam detection model has not been disclosed, and it remains unclear if it will be integrated as a default feature in future Android operating system updates. Both Google and GovTech indicated ongoing work and potential for expanding similar technology to other regions.