Meta has started testing an AI-powered feature on WhatsApp that can warn users about possible scams before they become more involved in a fraudulent conversation.
The optional feature, called Scam Alert, is being tested with a limited number of beta users, according to The Verge. It displays a warning banner inside a chat when the system detects patterns commonly associated with scams.
Unlike other AI-powered moderation tools, Scam Alert runs its machine-learning system on the user’s device. It does not upload message content to Meta’s servers for analysis.
In an engineering announcement Wednesday, Meta said the system “complements end-to-end encryption” while providing users with an optional warning when the model believes a conversation is likely to be a scam.
Message content is not uploaded for classification unless the user specifically chooses to share it.
When a possible scam is detected, a notification appears that can be seen only by the person receiving the message, not the other person in the chat. The user can then block the contact, report the account or continue the conversation.
If the warning is shown by mistake, users can mark the chat as trustworthy. This removes the warning and prevents Scam Alert from showing the same notification again. Users who want to help improve the model can also choose to submit the last five messages from the chat.
Scam Alert builds on another system WhatsApp introduced earlier this year to detect suspicious requests to link new devices to an account. Scammers commonly use this method to take control of accounts without needing a password.
Meta also said it removed more than 159 million scam-related ads last year, with 92% removed before users had a chance to report them. The company also removed 10.9 million accounts across Facebook and Instagram that were linked to criminal scam operations.
