Google DeepMind and Singapore Government announce new AI tool to protect citizens against scams
deepmind ·
Google DeepMind and the Singapore Government have announced a new on-device AI tool designed to protect citizens from scams while preserving privacy. Scams account for over half of reported crimes in Singapore, causing more than S$4 billion in losses since 2019, with complex social engineering tactics now driving up to 80% of these financial damages. To address this, Google DeepMind fine-tuned its lightweight Gemma model on local scam conversations, enabling it to distinguish between benign chatter and manipulative intent specific to the region. Unlike traditional server-side scanning, this solution processes data directly on smartphones, which is crucial for maintaining security within end-to-end encrypted messaging platforms. The system employs a two-stage pipeline that first screens messages efficiently and then performs contextual evaluation to confirm threats without draining battery life. Strict governance measures were implemented during development, including using a sovereign sandbox and excluding any personally identifiable information from training datasets. This collaboration demonstrates how governments and technology providers can work together to build public safety tools that do not compromise user privacy.
deepmind