Google and Singapore unveil private smartphone scam detector

Google and Singapore's Government Technology Agency have collaborated to develop an artificial intelligence system for detecting scams in private smartphone conversations. The system, announced at Singapore International Cyber Week, uses a modified version of Google's Gemma model to identify manipulation in conversations and warn users before they transfer money or share sensitive information.

The technology is intended to be integrated into messaging platforms and operating systems rather than released as a standalone application. The AI model undergoes two stages of assessment: a lightweight classifier initially checks messages for potential threats, followed by a more detailed analysis by the Gemma model for suspicious messages.

Developing the model involved training it on examples of scams and regional fraud patterns to make fraudulent exchanges appear credible to Singaporean users. The system operates locally on smartphones, analyzing conversations without uploading them to external servers, which is crucial for messaging services that use end-to-end encryption.

The partners emphasize that the model was trained using anonymised datasets and excludes any sensitive information. While the partners have not disclosed accuracy figures or independent testing results, the initiative aims to protect users from fraud schemes that rely on prolonged interaction, often accounting for up to 80% of scam losses in Singapore.