Google DeepMind and Singapore GovTech Deploy On-Device AI to Block Scams
A new collaboration uses lightweight Gemma models on user devices to detect conversational scams without sending chat data to the cloud.
Key points
- Google DeepMind and GovTech Singapore announced a new tool to protect citizens from scams.
- The system uses lightweight Gemma models running directly on user devices.
- The solution intercepts conversational threats in private chats with zero cloud data egress.
Google DeepMind and GovTech Singapore have announced a new artificial intelligence tool designed to protect citizens against scams. According to first-hand research published by Google DeepMind, the initiative leverages on-device processing to identify fraudulent interactions without compromising user privacy or sending data off the device.
The core of this solution involves lightweight Gemma models that run locally on users' hardware. These models are specifically tuned to intercept conversational scams within private chats. By keeping the processing on the device, the system ensures there is zero cloud data egress, meaning sensitive conversation details never leave the user's phone or computer for analysis.
Technical Approach and Privacy Design
The technical architecture prioritizes data minimization by avoiding cloud-based inference for sensitive content. Google DeepMind states that the use of lightweight Gemma models allows for real-time detection while maintaining strict privacy boundaries. This approach contrasts with traditional cloud-based AI security tools that often require uploading data for processing, which can raise privacy concerns among users and regulators.
The focus on on-device execution addresses the growing demand for privacy-preserving security measures. By processing data locally, the system reduces the attack surface associated with data transmission. It also ensures that the scam detection capabilities remain available even if network connectivity is intermittent, as the models do not rely on constant cloud communication for their primary function.
Target Audience and Implementation
The primary beneficiaries of this tool are citizens in Singapore, as the project is a joint effort with GovTech Singapore. The goal is to provide a scalable defense against the rising tide of digital fraud targeting individuals. The implementation focuses on conversational scams, which are increasingly sophisticated and often occur in private messaging applications where traditional security measures may not be effective.
This announcement follows Google DeepMind's continued expansion into applied AI domains. Earlier, in October 2026, the company released Gemini 4 Argon for robotics control, highlighting its focus on embodied reasoning. As Threat Times reported on 2026-10-13, that model scored 87 percent on the Embodied Reasoning benchmark, enabling robots to practice tasks in simulation before executing them in the physical world. The new scam detection tool represents a shift toward consumer-facing security applications.
What should you do?
Security professionals should monitor the rollout of this on-device scam detection tool for potential integration into broader enterprise security frameworks. Organizations operating in Singapore may consider how this technology aligns with their own fraud prevention strategies. Users should stay informed about the availability of the tool and understand its privacy implications before adoption.
What to do and how to stay safe: Google DeepMind
- Review current scam detection policies to see if on-device AI solutions offer better privacy compliance than cloud-based alternatives.
- Educate users on the differences between on-device and cloud-based security tools, emphasizing the data privacy benefits of local processing.
- Monitor official announcements from GovTech Singapore regarding the public release schedule and device compatibility for the Gemma-based tool.
- Assess existing private chat security measures to identify gaps that conversational AI detection could address without requiring additional data transmission.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Does this tool send my private chat messages to the cloud?
No, the system uses on-device Gemma models to process data locally, ensuring zero cloud data egress for private chats.
Who is developing this scam detection tool?
The tool is a joint development between Google DeepMind and GovTech Singapore.
What type of scams does this tool detect?
It is designed to intercept conversational scams that occur within private messaging chats.
