WhatsApp's new AI scam alert flags fraud before you reply — and it never leaves your phone

Meta is rolling out a new WhatsApp feature called Scam Alert that uses artificial intelligence to detect potentially fraudulent messages before users engage with them. The system runs entirely on the device and is designed to warn about risk as soon as a suspicious conversation appears.

The underlying model was trained on patterns found in chats with signs of fraud that users had reported to the company. It assesses the likelihood of a scam based on conversation structure and linguistic cues. When the algorithm flags a message as a possible fraud attempt, a warning appears in the chat, visible only to the recipient. From there, the user can choose to block the contact, report the issue, or continue the conversation.

If the user believes the message was mistakenly flagged, they can add the chat to a trusted list, after which the system will no longer check it. Meta also offers an optional way to help improve accuracy: users can voluntarily send the last five received messages to WhatsApp for review. The company stresses that no message content leaves the device for classification while Scam Alert is enabled, and no automatic data sharing with Meta or third parties takes place.

The feature can be toggled on and off in settings and is currently available as a limited beta. According to the U.S. Federal Trade Commission, WhatsApp users lost $425 million to scams in 2025 alone — a figure that highlights why Meta is pushing proactive fraud detection inside the app.