Whatsapp Tests on Device ML for Scam Detection with Privacy Preserving Analytics
WhatsApp is testing Scam Alert in limited beta, using on device machine learning to detect potential scam messages from non contacts. Meta's architecture keeps message content on the device while using confidential computing, Oblivious HTTP, differential privacy, and model transparency to measure performance and protect model delivery. By Leela Kumili
Meta is testing an optional Scam Alert feature for WhatsApp that uses on-device machine learning to warn users about messages that may be scams, without collecting message content. The feature relies on privacy-preserving techniques like confidential computing and differential privacy. WhatsApp trains the model on patterns seen in scam conversations reported by users.
If the model flags a message as likely scam, the user sees a warning but the sender does not. Users can mark chats as trusted to avoid future scam alerts. They can also share the last 5 messages with WhatsApp to improve the feature. Meta separates model training from performance measurement to enhance security. Model distribution is treated as a security boundary and the company publishes model details to a transparency ledger before deployment.
Written by urgent.news from InfoQ's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
