WhatsApp is testing a new optional feature called Scam Alert that uses a machine learning model running entirely on your device to warn you when an incoming message looks like a scam attempt. The feature is currently in limited beta with Bug Bounty researchers and is designed to catch messages from unknown contacts that match patterns associated with fraud — things like impersonation, urgent money requests, and romance or investment scams. When the model flags a conversation as a likely scam, you’ll see a warning in the chat with options to block, report, or continue talking. If you decide the warning was a mistake, you can mark the contact as trusted, which stops Scam Alert from flagging that chat again. WhatsApp says that if you do mark a chat as trusted, you can also optionally share the last five messages with WhatsApp to help improve accuracy — but that sharing is opt-in and only triggered by that choice.
The privacy angle here is a genuine selling point. The message content and the model itself never leave your phone; nothing is sent to Meta, WhatsApp, or any third party for classification. That means Scam Alert works alongside, rather than against, WhatsApp’s end-to-end encryption. The company has been building out a broader set of fraud protections lately: in March it added warnings for suspicious device-linking requests (a tactic used to hijack accounts by tricking users into scanning malicious QR codes), and earlier this year it launched “Strict Account Settings” aimed at journalists and activists at high risk of targeted spyware attacks. WhatsApp has more than 3 billion users across 180 countries, which means scammers target it heavily — this on-device approach is a practical step toward catching fraud attempts before users engage with them.