WhatsApp tests AI-driven Scam Alert to detect fraud

Meta is testing a new security feature for WhatsApp called ‘Scam Alert’ to combat digital fraud. Currently in limited beta, the tool uses on-device machine learning to identify suspicious patterns in messages, particularly those from unknown numbers.

To maintain user privacy and respect end-to-end encryption, the analysis is performed locally on the user's smartphone. This means message content is not sent to Meta or any external servers for scanning. The AI model is downloaded to the device once the user manually enables the feature in their settings.

When the system detects potential scam indicators—such as urgent requests for money, personal information, or suspicious links—it displays a warning banner within the chat. This alert is visible only to the recipient. Users can then choose to block the contact, report the conversation, or mark the chat as reliable to dismiss the warning. If a user marks a chat as reliable, they may optionally share the last five messages to help improve the model's accuracy.