DeepMind and Singapore build an on-device AI to spot scam chats

What happened

Google DeepMind and the Government Technology Agency of Singapore (GovTech Singapore) have fine-tuned an AI model to detect scams, working together under DeepMind's National AI Partnership with the Singapore Government ^[1]. The announcement was made on 14 October at Singapore International Cyber Week ^[1][2].

The model is built to run directly on a phone rather than on a remote server. When it spots manipulative patterns in a conversation, it triggers a real-time warning telling the user to pause before sharing sensitive information or money ^[1].

Why scams are the target

DeepMind says scams such as investment fraud, fake job offers and impersonation of public officials make up over half of Singapore's reported crimes, and have caused more than S$4 billion in losses since 2019 ^[1]. It adds that scams based on convincing conversations, rather than simple phishing, now account for up to 80% of all losses ^[1].

How it works

The model is based on Google's lightweight Gemma model, fine-tuned on local scam conversations, including impersonation of public institutions, banks and local figures ^[1]. DeepMind says generic models often struggle with local deception techniques ^[1].

To save battery, it uses two stages. A lightweight classifier first screens incoming messages. A fine-tuned Gemma model then analyses a larger part of the conversation to confirm a threat or dismiss a false alarm ^[1].

Privacy claims

Privacy is central to the pitch. DeepMind says that when deployed, the model will only access on-device conversations with the user's consent, under the terms of the app using it. Because it runs entirely on the device, "neither the Singapore Government nor Google, will be able to access users' conversations through its deployment" ^[1].

Training was done in a GovTech-managed Google Cloud environment. DeepMind says it excluded private or personally identifiable information and used anonymised datasets from scam reports and intelligence feeds. The model was tested against synthetic dialogues and local datasets ^[1].

Mark Chen, GovTech's director of trust and safety, said the approach lets the partners "catch highly targeted, local scam attempts in real time while keeping personal communications private" ^[1].

What we still do not know

The Straits Times reports that no release date has been announced and that it is unclear whether the model will be a default Android feature ^[2]. DeepMind says rollout in apps will be phased and developed with the Singapore Government ^[1]. It has not published figures for how accurate the model is.

Why it matters

Scam losses are a problem for banks, telecoms firms and ordinary customers alike. Many fraud tools look at transactions or scan messages in the cloud. This design looks at the conversation before any payment is made, while aiming to keep the content on the device.

DeepMind says it is exploring similar technology for other countries ^[1]. Businesses that handle customer messages should watch whether on-device checks can deliver protection without raising new privacy worries.

Source: Google DeepMind — Google DeepMind and Singapore Government collaborate to build fine-tuned AI model for scam detection