Google DeepMind 与新加坡 GovTech 推出基于 Gemma 微调的手机端诈骗检测模型

October 14, 2026 Responsibility & Safety

JJ Geewax and Miguel de Andrés

The collaboration delivers an on-device scam detection model that helps protect citizens without compromising privacy.

Scams make up over half of Singapore’s reported crimes, and account for more than S$4 billion in losses since 2019. At the same time, scam tactics have shifted from simple phishing to complex social engineering, whereby attackers manipulate victims with convincing conversations before requesting money. Such scams now account for up to 80% of all losses.

To address this, Google DeepMind and the Government Technology Agency of Singapore (GovTech) have been working together as part of our National AI Partnership with the Singapore Government. This has allowed us to deliver an on-device scam detection model - optimized to run on typical smartphones in Singapore - to tackle an urgent safety challenge while preserving citizens' privacy.

On-device AI protection without compromising privacy

Generic models often struggle with localised deception techniques, making scam detection harder. To address this, Google DeepMind's lightweight Gemma model was fine-tuned on local scam conversations. These ranged from impersonations of public institutions, banks, and local figures, to trending regional schemes, ensuring the model can distinguish benign social chatter from genuine manipulation.

This new model also processes conversation data on the device. Traditionally, threat detection relies on server-side scanning of messages in the cloud, but many messaging platforms are end-to-end encrypted - meaning they cannot inspect private conversations. People need a way to stay safe from fraud at the same time as keeping conversations private, and our solution is to bring the threat analysis to the device. This allows faster analysis of conversation patterns to detect manipulative intent and allow citizens to act before financial or credential exposure can occur.

Under the hood: local context and edge efficiency

When the model detects manipulative conversational patterns, it triggers a real-time warning directly on the user's device advising them to pause and exercise caution before sharing sensitive information or funds.

Running a complex scam classifier model on modern smartphones is no easy task, with battery efficiency being a critical area of concern. To do this we're relying on a two-stage pipeline:

• Stage 1: A lightweight, efficient classifier screens inbound messages, using minimal power, focused on catching potential threats.
• Stage 2: Contextual evaluation with our fine-tuned Gemma model, where we analyze a larger snapshot of the conversation to confirm a threat or discard a false alarm.

Built on strict governance

Presented at Singapore International Cyber Week (SICW 2026), this collaboration provides a model for messaging platforms balancing user protection with end-to-end encryption. Strict governance was maintained throughout development:

Sovereign Sandbox: Fine-tuning took place within GovTech's managed and maintained GCP environment, keeping training data isolated.

No Citizen Data: Training strictly excluded private or personally identifiable information, utilising anonymised datasets derived from scam reports and intelligence feeds.

Stress-Testing: The model was tested against synthetic dialogues and local corpora to cover diverse scam categories and evolving impersonation tactics.

The future of on-device safety

The project demonstrates how AI can help messaging services counter criminality and shows safety nor privacy need be compromised by running protective AI directly on-device. It also shows how governments and model providers can work together to build tech for public good. Messaging platforms and operating system vendors can directly integrate these lightweight models to safeguard users locally, ensuring chat interactions remain secure without exposing private data.