Meta deploys on-device AI to curb WhatsApp fraud

The Scam Alert feature, currently in limited beta, triggers a notification if a message pattern mirrors known fraudulent tactics. These warnings remain private, visible only to the recipient, who then decides whether to block, report, or mark the sender as trusted. By designating a contact as trusted, users can override the system and optionally share the five most recent messages to refine the model's accuracy.

Meta emphasizes that the classification process occurs entirely on the user’s hardware. No message content is transmitted to Meta’s servers, and no automated reporting occurs without explicit user input. This privacy-first approach arrives as the platform grapples with a surge in sophisticated schemes, including wire transfer fraud and pig butchering scams, which have contributed to a staggering $2.1 billion in total social media-based losses this year.