WhatsApp deploys on-device AI to flag scam messages in real time

Meta is rolling out a new on-device artificial intelligence feature designed to identify scam attempts on WhatsApp before users fall victim. The Scam Alert tool, now entering limited beta testing, analyzes incoming messages using machine learning that runs entirely on the user's phone.

When the AI model flags a message as a potential scam, WhatsApp displays a warning visible only to the recipient—not to the sender. Users then choose whether to block the contact, report it, or continue the conversation despite the alert.

The feature includes a feedback mechanism: if users believe a warning was incorrectly triggered, they can mark the chat as trusted. This removes the alert and prevents future flags for that conversation. Users who mark a chat as trusted can optionally share the last five received messages with WhatsApp to help refine the model's accuracy.

Why it matters

WhatsApp's massive user base—more than 3 billion people—makes it a prime target for fraud schemes ranging from wire transfer scams to elaborate pig butchering operations. The Federal Trade Commission reported that victims lost $425 million to scams via WhatsApp alone in 2025, part of more than $2.1 billion lost across all social media platforms. An on-device detection system that preserves end-to-end encryption while still providing protection represents a technical approach other messaging platforms may need to adopt as scam sophistication increases.

Privacy-first architecture

Meta emphasized that the scam detection system processes all data locally. According to the company, "no message content leaves the user device for classification" when Scam Alert is active. Nothing gets automatically reported to WhatsApp, Meta, or third parties. Users retain full control and can disable Scam Alert at any time.

This architecture addresses a longstanding tension in encrypted messaging: how to provide safety features without compromising the privacy guarantees that make services like WhatsApp attractive to users concerned about surveillance or data collection.

Expanding fraud defenses

The Scam Alert feature builds on fraud prevention tools Meta introduced earlier in 2026. The company previously launched scam detection specifically for device linking requests—a common attack vector where scammers attempt to hijack accounts by tricking users into authorizing access from a new device.

Together, these features represent Meta's attempt to address fraud systematically across multiple threat vectors while maintaining its commitment to end-to-end encryption, a balance that has drawn scrutiny from regulators and law enforcement agencies seeking greater platform accountability for harmful content.

The details were first reported by The Verge.