WhatsApp is testing (and has begun a limited beta rollout of) “Scam Alert,” an optional on-device AI feature that warns users about potential scam messages from unknown/non-contacts while preserving end-to-end encryption
Key details
- How it works: When enabled, the app downloads a machine-learning model to your device. The model scans incoming messages from people not in your contacts and looks for patterns linked to known scams (conversational structure and linguistic signals, trained on previously user-reported scam conversations). It runs entirely on-device—no message content is sent to WhatsApp, Meta, or any third party for classification or automatic reporting.
- What users see: If a message is flagged as a likely scam, a private warning appears only in your chat (the sender cannot see it). You then choose to block/report the contact, continue the conversation, or mark the chat as trusted (which removes the warning and stops future flags for that chat). Optionally, if you mark a chat as trusted, you can share the last 5 messages to help improve the model.
- Privacy & control: It is optional and off by default—you must turn it on manually in settings. You can disable it anytime. End-to-end encryption remains fully intact. Meta emphasizes verifiability measures (e.g., transparency logs for model versions and scanned activity that users can review, plus third-party logging of model releases) and is expanding its bug-bounty program to cover the feature.
- Status: First spotted in Android beta code earlier in 2026. As of mid-August 2026, Meta published a detailed technical overview and started a limited beta rollout in some regions. A wider release is expected after further testing; no exact full-rollout date has been announced.