SAN FRANCISCO — Meta Platforms is rolling out an optional Scam Alert feature on WhatsApp, using on-device machine learning to flag suspicious messages. The Federal Trade Commission reported victims lost $425 million to WhatsApp scams in 2025. The feature is currently in limited beta.
When the AI model identifies a likely scam attempt, a warning appears in the user's chat, visible only to that user. Users can block, report or continue the exchange.
Users can mark a chat as trusted if a warning is incorrect, removing the alert and preventing future flags for that conversation. To improve accuracy, users can opt to share the last five messages received with WhatsApp after marking a chat as trusted.
Meta said no message content leaves the user's device for classification. Nothing is auto-reported to WhatsApp, Meta or any other entity without explicit user action. Users can disable Scam Alert at any time.
WhatsApp's 3 billion users make it a large platform for scam activity. Common schemes include wire transfer fraud and pig butchering scams. The $425 million lost on WhatsApp in 2025 was part of a broader $2.1 billion lost to social media scams that year.
Meta's rollout follows comparable moves by other technology companies. Google Pixel phones and Samsung Galaxy S26 series devices introduced similar on-device scam detection features in February 2026.
Reduced scam activity can drive higher user engagement and retention. Deploying AI models directly on user devices, processing data locally, also addresses privacy concerns that have historically complicated cloud-based content analysis — a real competitive consideration as messaging apps compete on trust as much as features.
Beyond device manufacturers, specialized applications such as Scamless and Charley from Charlemagne Labs also use AI to monitor incoming messages and warn users of potential fraud, reflecting a broader industry push toward client-side AI for security.
The limited beta gives Meta data to refine the Scam Alert model's accuracy before a wider launch. The user feedback loop — including the option to share message data — is central to improving the AI models over time.
