Models · 2026-10-14

Google DeepMind and the Singapore government are building AI that will look for scams in private chats

Google DeepMind together with Singapore's state technology agency GovTech is training lightweight Gemma AI models to recognize scam messages directly on a phone, without sending data to the cloud.

✨ What is happening

Google DeepMind together with the Singaporean state agency GovTech Singapore is developing an AI model designed to help people avoid falling for scammers in messaging. The focus is on so-called conversational scams, when a fraudster gradually gains a person's trust in a chat and then asks for money or personal data.

The project uses Google's Gemma models. These are relatively lightweight AI models that can run directly on a device, for example on a smartphone.

📌 Why it matters

The main feature of the project is that the model works on the device itself. According to the authors, messages from private chats are not transferred to external servers, meaning personal correspondence stays with the person.

For ordinary users, this means a potentially lower risk that scammers will manage to deceive them, while also preserving privacy. Detailed results of the work and launch dates are not specified in the material.

Sources: 🔗 Google DeepMind Blog