Meta announced an optional on-device AI feature called Scam Alert for WhatsApp to automatically flag suspicious messages [1, 2].
The tool aims to reduce fraud and protect users from scam attacks without compromising the platform's commitment to privacy. By processing data locally, the system attempts to balance security with the preservation of end-to-end encryption [1, 3].
The feature was announced on Aug. 14, 2026 [1]. It is currently being rolled out in a beta testing phase to users worldwide [1, 2].
Unlike cloud-based AI systems, Scam Alert operates directly on the user's mobile device [1, 2]. This means the AI analyzes the content of the chats locally to identify patterns associated with fraud, ensuring that the messages remain encrypted and are not read by Meta's servers [2].
Users who opt into the feature will receive notifications when the AI detects a potential scam [1, 3]. The system is designed to help users detect and prevent attacks in real time, providing a layer of defense against increasingly sophisticated social engineering tactics [1, 3].
Meta said the tool is intended to reduce the volume of successful scams on the platform [3]. The company is testing the feature's efficacy in various regions before a full global release [1, 2].
“The system attempts to balance security with the preservation of end-to-end encryption.”
This move signals a shift toward 'edge AI,' where complex security processing happens on the hardware rather than the cloud. By keeping the analysis on-device, Meta addresses the primary criticism of AI moderation in encrypted spaces: the trade-off between safety and privacy. If successful, this model could become the standard for other encrypted messaging services facing similar pressure to combat fraud.
