WhatsApp Begins Limited Test of AI Scam Alerts for Unknown Senders

WhatsApp is testing a new AI-powered Scam Alert feature designed to flag suspicious messages from unknown senders while preserving end-to-end encryption. The optional feature, currently in a limited beta, uses on-device machine learning to identify potential scams based on linguistic and conversational patterns. If the model detects a likely scam, WhatsApp displays a warning visible only to the recipient.

Users can then choose to block or report the sender, continue the conversation, or mark the chat as trustworthy. If the alert is incorrect, marking the chat as trusted will remove the notification and prevent future alerts for that chat. Users can also voluntarily share recent messages with WhatsApp to help improve the model. Meta asserts that Scam Alert runs locally on devices and does not access users' encrypted conversations.

The detection model and any model versions are recorded on a third-party transparency ledger, and devices verify signatures and file hashes before loading a model. WhatsApp collects limited performance data, which is aggregated and protected using confidential computing and differential privacy. Scammers are increasingly using messaging platforms to build trust, pressure victims, and solicit money or personal information.

WhatsApp aims to add an automated layer of protection without inspecting private conversations. However, Scam Alert is not a guarantee that a message is safe or fraudulent, as machine learning can miss sophisticated scams or incorrectly flag legitimate conversations. The feature remains an experiment, and Meta plans to continue testing, expand bug-bounty coverage, and refine the model before deciding on a broader rollout.