Google DeepMind and the Government Technology Agency of Singapore (GovTech) have developed a scam-detection model that runs on smartphones and analyzes conversations without sending message content to external servers.

In a statement on Wednesday, Google DeepMind said the project forms part of Google’s National AI Partnership with the Singapore government.

Scams account for more than half of Singapore’s reported crimes, and losses exceeded S$4 billion since 2019. Social engineering, in which attackers build trust through convincing conversations before asking for money, now accounts for up to 80 percent of those losses.

The model is based on Google’s lightweight Gemma model, which the partners fine-tuned on local scam conversations. The training material covered impersonation of public institutions, banks and local figures, along with emerging regional schemes.

The model works in two stages. A lightweight classifier screens incoming messages at low power cost and flags suspicious ones. The fine-tuned Gemma model then analyzes a larger slice of the conversation to confirm a manipulative pattern or dismiss a false alarm.

When it confirms a threat, it displays a real-time warning on the device, advising the user to pause before sharing information or money.

The partners said the model can access conversations only with user consent and in line with each application’s terms of use. Neither the government nor Google can read conversations through its deployment.

Google said the rollout will be phased and developed with the Singapore government. The model would be integrated into applications used by citizens.

Fine-tuning took place in a GovTech-managed cloud environment, with training data kept separate. The partners said training excluded private and personally identifiable information, using anonymized data from scam reports and intelligence feeds.

Google DeepMind said the model was tested against synthetic dialogues and local language patterns across several scam categories.

Google Cloud opens Singapore Engineering Center to build AI and cloud products for global markets