WhatsApp is now using AI to catch scammers before they can trick you
WhatsApp is looking to tackle the increasing number of scams on the platform by using the power of artificial intelligence. The company on Thursday announced a new experimental feature called Scam Detection that uses on-device AI models to flag suspicious conversations.
“As scam tactics evolve, from impersonation to social engineering to AI-generated lures, we’re always evolving as well, so that our protections stay ahead of scammers while protecting people’s personal messages with end-to-end encryption,” Meta wrote in a blog post.
As per the US FTC, via The Verge, WhatsApp was the second-highest source of fraud contact methods in 2025. It noted that of the $2.1 billion worth of scams perpetrated through social media, $4.25 billion belonged to WhatsApp alone.
How does Scam Alert on WhatsApp work?
Once the feature has been turned on, WhatsApp downloads a machine learning model on the user’s device. This model will then run inferences on the user’s phone to identify if an incoming message from a non-contact matches known scam patterns.
WhatsApp says this on-device AI model is trained on “patterns observed in scam conversations from reports that users have sent to us.”
The model is said to perform probabilistic classification based on conversational structure and linguistic signals.
If the model decides that the message is a likely attempt to scam users, it will start showing them a warning that will not be visible to the other person.
Users will also have the option to either block the person, report them or continue the conversation. They can also mark the chat as trusted, and WhatsApp will remove the warning. The company says Scam Alert will not flag that chat in the future.
“If a user marks that they trust a chat, they can also opt in to share the last 5 messages received with WhatsApp to help improve the feature’s accuracy,” Meta noted.
The social media behemoth also notes that no user data is automatically shared with Meta, WhatsApp or any other third party during this process.
“All inference happens on-device and no message content leaves the user device for classification. The minimal telemetry needed to measure whether the feature is working,” Meta wrote.
The company also added that Meta or WhatsApp will not deliver a specific model to a specific user and every model version will be published on a public transparency ledger. Meta has also committed to publishing its model weights so that independent security researchers can verify that these models have been built for scam detection only.
