WhatsApp's new 'Scam Alert' feature employs a local machine learning model trained on user-reported scam conversations to detect potential scam messages from non-contacts. The model uses linguistic signals and probabilistic classification based on conversational structure to identify likely scams. Importantly, all processing occurs on the user's device, with no message content or model data leaving the device, maintaining end-to-end encryption and user privacy. Users can mark chats as trusted to prevent future warnings and optionally share limited message data to improve model accuracy. This feature is part of a broader security effort including prior updates that warn users about fraudulent device-linking requests and lockdown-style account protections for high-risk individuals.

The feature enhances user protection against scam messages by providing real-time warnings on potentially fraudulent communications without compromising message privacy. It reduces the risk of users falling victim to scams by enabling informed decisions to block or report suspicious contacts. Since the feature is optional and local to the device, it does not introduce new privacy risks or data exposure. The rollout is currently limited and in beta, so broad impact depends on adoption and further refinement.

The Scam Alert feature is an optional security enhancement that users can enable or disable at will. Since it is currently in limited beta rollout, users interested in additional scam protection can opt in to participate. No other immediate action is required. Users should continue to exercise caution with messages from unknown contacts. WhatsApp manages the deployment and updates of this feature; no manual patching is necessary.