Use on-device AI to detect scams in private chats without sending data to the cloud
New system uses lightweight AI models on phones to detect fraud in private chats without sending data to the cloud.
Key points
- Google DeepMind and GovTech Singapore collaborated to create a scam detection tool.
- The system runs lightweight Gemma models directly on user devices.
- Conversational data stays on the phone with zero egress to the cloud.
Google DeepMind and the Singapore government have launched a new system that detects scams on mobile phones without sending private chat data to the cloud. This collaboration aims to protect users from fraud while preserving their privacy by keeping all processing local to the device.
In plain English
Most AI tools require sending your data to powerful servers for analysis. This new approach changes that model. Instead of uploading your messages, the AI runs directly on your phone. This means your private conversations never leave your device, which protects your personal information while still allowing the system to identify suspicious activity in real time.
The background
According to research published by Google DeepMind, the partnership between the tech giant and GovTech Singapore focuses on building fine-tuned AI models specifically for scam detection. The project utilizes lightweight versions of the Gemma model family. These smaller models are designed to operate efficiently on consumer hardware, allowing for rapid analysis of text messages and chat logs without the latency associated with cloud-based processing.
What changes now
Users in Singapore can now benefit from proactive scam detection that respects data privacy standards. The system intercepts conversational scams within private chats. Because the models are fine-tuned for this specific task, they can recognize patterns associated with fraud attempts. The key technical achievement is the elimination of cloud data egress, ensuring that sensitive personal communications remain strictly on the user's device throughout the detection process.
Key terms explained
On-device AI refers to artificial intelligence models that run locally on a smartphone or computer rather than in remote data centers. This approach enhances privacy because user data does not travel over the internet. Fine-tuning is the process of adapting a general AI model to perform a specific task, such as identifying scam language. Lightweight models are smaller versions of AI that require less processing power, making them suitable for mobile devices.
What to do and how to stay safe: Google DeepMind
- Review your device settings to ensure you are using the latest security updates provided by your manufacturer.
- Monitor your notifications for any new scam detection features enabled by your mobile carrier or operating system.
- Verify that your messaging apps have permission to run background security services if such options are available.
- Stay informed about new privacy features that keep your data local rather than sending it to external servers.
General security guidance from the Vector Update newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Does this system send my private messages to Google or the government?
No, the system processes data on your device with zero cloud data egress, keeping your chats private.
What AI model is used for this scam detection?
The collaboration uses lightweight on-device Gemma models fine-tuned for scam detection.
Who developed this scam detection technology?
It was developed through a collaboration between Google DeepMind and GovTech Singapore.
