Meta is testing a new WhatsApp feature that uses on-device AI to spot scam conversations before you fall for them. It’s called Scam Alert, it’s optional, and right now it’s only live for a limited group of beta testers, having started rolling out around mid-August 2026.
Here’s the mechanic: once someone turns it on, WhatsApp downloads a small machine-learning model straight to the phone. That model watches messages coming from people who aren’t saved as contacts and looks for patterns scammers tend to use, things like manufactured urgency, requests for personal or financial details, impersonation attempts, and the slow trust-building common in romance scams. Rather than scanning for specific keywords, the system is reading the shape of the conversation itself, the tone and structure of how it develops.
When it spots something that looks off, a warning banner shows up right inside the chat. From there you can block the sender, report the account, or just keep talking if you’re confident it’s fine. Get a false alarm? Mark the chat as trusted and the warning won’t fire again for that conversation.
The privacy angle is the part Meta keeps emphasizing. Because the model runs locally, message content never has to leave the device to get classified, and nothing about a flagged chat gets sent back to WhatsApp automatically. In its August 12 engineering announcement, Meta put it plainly: no message content leaves the device for classification. If a user wants to help improve the model after marking a chat trusted, they can choose to share the last five messages they received, but that’s a manual opt-in, not something that happens by default.
Meta says it’s also publishing details about the underlying models and making the weights available so outside security researchers can pick them apart, and it’s working with WhatsApp’s bug bounty community to stress-test the system before it goes any further. There’s no announced date for a broader rollout yet.
This isn’t Meta’s first swing at the problem. Earlier this year the company introduced an automated system that flags suspicious linked-device requests on WhatsApp, and Scam Alert reads as the next layer on top of that, aimed at conversations rather than account access. Given how much scam activity already runs through WhatsApp, particularly payment fraud and fake-identity schemes in markets like India, a feature that catches the con before someone actually sends money is a fairly obvious thing to want. Whether it works well enough to leave beta, and whether it starts flagging perfectly normal messages from unfamiliar businesses, is still an open question.
For Latest Tech Updates Please follow us :
facebook | twitter | instagram | telegram | whatsapp
