- Meta has been testing a Machine Learning and AI-based 'Scam Alert' feature that works on-device and warns users of potential cyber threats, and has shared a preview.
- Before the official public launch, WhatsApp will showcase its early technical overview and make the feature available to a select group of testers via WhatsApp Beta testing in the coming days to seek feedback.
- The Scam Alert is an optional feature that runs an on-device machine learning model to alert users about potential scam messages, and no message content leaves the device for classification or is auto-reported.
- The feature complements end-to-end encryption while providing a user-controlled, optional scam alert when the model believes there’s likely a scam.
- Once turned on, the Scam Alert feature downloads a machine learning model to the device, where it runs inferences to classify whether incoming messages from non-contacts match known scam patterns.
- If the model detects a likely scam, the user sees a warning in the chat (not visible to the other person) and can decide to block, report, or continue the conversation.
- Users can mark a chat as trusted if a warning is incorrectly flagged, which removes the warning and prevents future flags for that chat; they can also opt in to share the last 5 messages to improve accuracy.
- Meta will continue to run a dedicated Bug Bounty community to stress-test WhatsApp and weed out security vulnerabilities to ensure user privacy protection.
- The model is trained on patterns from scam conversations reported by users, performs probabilistic classification based on conversational structure and linguistic signals, and no content is automatically reported to WhatsApp, Meta, or any third party.
Meta's new Scam Alert feature for WhatsApp aims to enhance user safety by utilizing on-device machine learning to detect potential scams without compromising privacy.1234
"WhatsApp is committed to helping people stay safe while protecting the privacy of their messages," the company stated, emphasizing the importance of evolving protections against sophisticated scam tactics.
The feature operates by downloading a machine learning model to the user's device, which analyzes incoming messages from non-contacts for known scam patterns.
No message content leaves the device for classification, ensuring that user privacy is maintained. If a potential scam is detected, users receive a warning in the chat, which remains invisible to the sender.
Users can choose to block, report, or continue the conversation, and if they trust a flagged chat, they can mark it as such, which removes the warning and helps improve the feature's accuracy by optionally sharing the last five messages received.
Before the public launch, WhatsApp is testing the feature with a select group of users and will continue to run a Bug Bounty program to identify and address security vulnerabilities.11
This initiative reflects Meta's ongoing commitment to user privacy while adapting to the evolving landscape of online scams.
“The optional feature runs a machine learning model on-device, so no message content leaves the device for classification or is auto-reported. Users can mark a chat as trusted if a warning is incorrect, and can opt in to share the last 5 messages to improve accuracy.”
Deccan Herald
