WhatsApp is adding another layer of protection against messaging scams, this time using machine learning that runs directly on a user’s phone. The new Scam Alert feature is entering a limited beta and is designed to identify conversations that show signs of fraudulent activity before users act on them.
The feature is optional and works by analysing incoming messages locally on the device. When WhatsApp’s model determines that a conversation may be suspicious, it displays a warning inside the chat. The person on the other end does not see the alert, leaving the recipient to decide whether to block the account, report it or continue talking.
Users can also tell WhatsApp when the system gets it wrong. Marking a conversation as trusted removes the warning and prevents Scam Alert from flagging that particular chat again. Users who do so can separately choose to send the five most recent received messages to WhatsApp to help improve detection accuracy.
That distinction matters because automated scam detection creates an obvious privacy question for an encrypted messaging platform. Meta says message content used for classification stays on the device and that suspicious conversations are not automatically reported to WhatsApp, Meta or another party. Scam Alert can also be disabled entirely.
The approach reflects a broader shift toward running safety-related AI models locally rather than processing private communications in the cloud. On-device models can provide automated analysis while limiting how much personal information needs to leave a smartphone, although their usefulness will depend heavily on how accurately they can distinguish genuine conversations from increasingly sophisticated scams.
WhatsApp has already been expanding its anti-fraud tools. Earlier in 2026, Meta introduced scam detection around suspicious device-linking requests, another technique attackers can use to gain access to accounts. The latest feature moves the defence deeper into conversations themselves, where social engineering can be harder to identify through conventional security checks.
Google has pursued a similar strategy with AI-powered scam detection in Google Messages, suggesting that messaging platforms increasingly see machine learning as a practical defence against fraud. The challenge is significant because scammers routinely change language, identities and tactics to avoid automated systems.
The financial impact provides some context for the investment. According to US Federal Trade Commission figures cited in the source material, consumers reported losing $425 million through scams involving WhatsApp in 2025. Reported losses associated with social media scams more broadly exceeded $2.1 billion during the year.
WhatsApp’s enormous user base — more than three billion people — inevitably makes it an attractive target for fraud ranging from impersonation and wire-transfer schemes to longer-running investment scams.
For now, the limited beta means Scam Alert is not yet a universal safeguard, and Meta has not eliminated the most difficult part of scam prevention: deciding when an unusual conversation is genuinely dangerous. Keeping that analysis on the phone, however, gives WhatsApp a way to experiment with AI-powered protection without requiring every private conversation to become cloud-processed security data.
