Meta introduces on-device AI alerts to flag WhatsApp scam messages

Meta has begun testing a Scam Alert option for WhatsApp that relies on an AI model running locally on users' devices to identify likely scam messages. When the model flags a chat, a warning appears only for the recipient, who can then block, report, or keep the conversation going. Incorrect warnings can be dismissed by marking the chat as trusted, which also stops further alerts for that conversation and optionally allows the user to share the last five messages to refine the model.

The company emphasizes that no message content leaves the device for classification and that no data is auto-reported to WhatsApp, Meta, or third parties. Users retain full control and may turn the feature off whenever they wish. WhatsApp, with over three billion users, has been a major target for scams such as wire-transfer fraud and pig-butchering schemes, which the FTC says cost victims $425 million in 2025 alone.