WhatsApp gets on-device AI to warn users of suspected scams

Meta is introducing Scam Alert, an optional safety feature for WhatsApp that leverages on-device machine learning to identify and warn users about suspicious messages. The tool, currently in limited beta, represents the latest effort to combat the wave of fraud plaguing the platform.

How it works#

The AI model runs directly on your phone, analyzing incoming messages to spot patterns typical of scams. When the system detects a likely fraudulent conversation, a warning appears in your chat—visible only to you—giving you the choice to block the sender, report them, or continue talking. If the warning was incorrect, you can mark the chat as trusted, which disables future alerts for that conversation.

Users who trust a chat can optionally share the last five messages they received with WhatsApp to help refine the detection system's accuracy over time. Importantly, Meta states that "no message content leaves the user device for classification" and nothing is automatically reported to Meta or WhatsApp unless you choose to share it. The feature can be disabled at any time.

The scale of the problem#

With over 3 billion users, WhatsApp has become a major vector for scam activity ranging from wire-transfer fraud to pig butchering schemes. The FTC reported that scam victims lost $425 million through WhatsApp alone in 2025, part of a broader $2.1 billion in social media scam losses that year.

Building on existing protections#

This addition follows Meta's earlier rollout of scam detection for device linking requests on WhatsApp, signaling a sustained push to make the messaging app a safer space as fraud tactics evolve.