WhatsApp is testing a new AI-powered "Scam Alert" feature designed to identify potentially fraudulent messages and warn users before they interact with suspicious conversations. The system runs directly on a user's device and analyses conversation patterns and linguistic cues associated with scams.
How the AI Model Was Trained
The model has been trained on patterns identified in conversations showing signs of fraud that were reported to the company through user complaints. Rather than checking isolated messages, it assesses the structure of an entire conversation, along with the language used, to determine the likelihood that a message may be part of an attempted scam.
When the system identifies a message as potentially fraudulent, a warning appears within the chat, visible only to that specific user. From there, the person can decide whether to block the other party, report the issue, or simply continue the conversation as they see fit.
Scam Detection Runs Directly on Your Device
Meta says the Scam Alert system performs its classification entirely on-device, meaning message content doesn't leave the user's phone or device for the detection process while the feature is enabled. The company says data isn't automatically sent to Meta or other organisations as part of this classification process, an approach specifically designed to allow suspicious conversations to be assessed while keeping the analysis local and private.
Users who believe a conversation has been incorrectly flagged can add that chat to a trusted list, after which the system will no longer check that particular conversation going forward. WhatsApp also provides an option for users to voluntarily send the last five messages received through the platform, helping improve the accuracy of the model over time based on real-world examples.
Users Stay in Control of What Happens Next
The warning is designed to give users an opportunity to pause and reconsider suspicious interactions before taking any further action. After receiving an alert, users retain full control over whether they block the sender, report the conversation, or continue communicating regardless. The feature can be toggled on or off through settings and is currently available as part of a limited beta test, meaning it isn't yet rolled out broadly to all users.
The system's focus on conversational patterns, rather than simply flagging isolated messages, allows it to pick up on linguistic cues and the way a suspicious interaction actually develops over time, which matters a lot given how many scams unfold gradually rather than in a single obvious message.
The Scale of the Problem This Feature Is Trying to Address
This feature comes amid substantial losses linked to online fraud globally. According to figures attributed to the US Federal Trade Commission, WhatsApp users lost 425 million dollars to fraudulent schemes in 2025 alone, a figure described as part of total social media fraud losses exceeding 2.1 billion dollars.
Fake Money Transfers and "Pig Butchering" Scams Among the Key Threats
Among the common fraud methods cited are schemes involving fake money transfers and so-called "pig butchering" scams. These scams typically involve fraudsters gradually building trust with potential victims over an extended period, before attempting to persuade them to transfer money, often for fake investment opportunities. This makes the development of a conversation itself an important potential indicator of fraudulent activity, which is exactly why WhatsApp's new feature analyses conversational structure rather than just isolated red-flag words.
By analysing this conversational structure and linguistic signals directly on the device, WhatsApp's Scam Alert feature is designed to catch warning signs before users proceed further into potentially fraudulent interactions. The limited beta test will offer an opportunity to assess how effective the system actually is in practice, while users who encounter incorrect warnings can simply mark conversations as trusted to avoid future false alerts. Overall, the feature represents a fairly notable attempt to use artificial intelligence to identify scam-related behaviour, while deliberately keeping automated message classification confined to the user's own device rather than sending data elsewhere.
FAQs
Q1. How does WhatsApp's new Scam Alert feature actually work?
The AI-powered system analyses conversation patterns and linguistic cues directly on the user's device, flagging potentially fraudulent messages with a warning visible only to that user, without sending message content off the device.
Q2. What can users do after receiving a scam alert warning?
Users can choose to block the sender, report the conversation, continue communicating, or mark the chat as trusted if they believe it was incorrectly flagged.
Q3. How much have users reportedly lost to fraud on WhatsApp?
According to figures attributed to the US Federal Trade Commission, WhatsApp users lost 425 million dollars to fraudulent schemes in 2025, part of over 2.1 billion dollars in total social media fraud losses.