Google DeepMind, Singapore Develop AI To Detect Scam Conversations

Singapore: Google DeepMind and Singapore’s Government Technology Agency (GovTech) have developed an artificial intelligence model designed to detect scam conversations on smartphones and warn users before they share money or sensitive information.

Announced during Singapore International Cyber Week on Wednesday, the technology aims to identify social-engineering tactics in which fraudsters impersonate government officials, bank representatives or public figures to manipulate victims.

The system uses a lightweight version of Google’s Gemma AI model, fine-tuned to recognise scam patterns specific to Singapore. It analyses conversations directly on the device, allowing suspicious behaviour to be detected without sending private messages to external servers.

The model uses a two-stage screening process. An initial, low-power classifier checks incoming messages for suspicious signals, while the AI model examines flagged conversations for broader patterns of manipulation. If a potential scam is identified, users receive an on-device warning advising them to pause before transferring money or disclosing confidential information.

The approach is designed to support privacy, particularly in end-to-end encrypted messaging environments. According to the developers, training relied on anonymised scam reports and intelligence feeds rather than private, personally identifiable citizen data.

The initiative builds on Singapore’s existing fraud-prevention efforts, including ScamShield. However, a public release date and details about integration with smartphones or messaging platforms have not yet been announced.

Subscribe