Google DeepMind brings on-device scam detection to Singapore
Google DeepMind and Singapore’s GovTech have built an on-device model that warns about scam conversations while keeping message analysis local.
Google DeepMind and Singapore’s Government Technology Agency (GovTech) have announced an on-device scam-detection model designed to warn people about manipulative conversations without sending private chats to the cloud. 1
How the protection works
The system uses a two-stage pipeline. A lightweight classifier first screens incoming messages for possible threats. Suspicious conversations then receive a broader contextual check from a Gemma model fine-tuned on anonymised local scam conversations. 1
When the model detects manipulative patterns, it can trigger a real-time warning on the device, advising the user to pause before sharing sensitive information or money. Google DeepMind says the approach is intended to work with messaging services that use end-to-end encryption, where conventional server-side scanning cannot inspect private conversations. 1
Built for Singapore’s scam landscape
The training work focused on localised deception, including impersonation of public institutions, banks and local figures. Google DeepMind says the model was developed inside GovTech’s managed Google Cloud environment, with training data kept isolated and no private or personally identifiable citizen data included. Testing used synthetic dialogues and local corpora to cover different scam categories. 1
The announcement places the project against Singapore’s substantial fraud problem: scams make up more than half of reported crimes and have caused more than S$4 billion in losses since 2019, according to Google DeepMind’s account. 1
What happens next
The project was presented at Singapore International Cyber Week 2026. Google DeepMind says the work could provide a model for messaging platforms and operating-system vendors that want to add local safety checks while keeping conversations private, but the announcement does not identify a consumer app or universal rollout date. 1
Our opinion
This is a more convincing privacy pitch than simply promising that AI will “make messaging safer”, because the useful detail is where the analysis happens: on the phone, with a lightweight first pass and a more focused second stage. The harder test will be false positives, battery use and how the warnings behave across different messaging apps. For now, the announcement is best read as a promising deployment pattern, not proof that every scam conversation can be detected reliably.
1 is the official Google DeepMind source.