WhatsApp is testing a security feature on Android that could give users an early warning when a message from an unknown contact appears suspicious. Called Scam Alert, the optional feature is based on an on-device machine-learning model to identify potentially fraudulent messages without sending their content to Meta or WhatsApp's servers.
It is currently available only to a fraction of Android beta testers. Such beta testing is a first step in WhatsApp’s journey to protect against online fraud and preserve the privacy of end-to-end encrypted chats.
Scam Alert brings fraud detection directly to the phone
Unlike other systems which rely on cloud servers to analyse message content, WhatsApp's Scam Alert is designed to first assess messages on the user's smartphone. The machine learning model scans messages that are not saved in the user's contacts and determines whether the conversation could potentially be a scam.
WhatsApp says that message content is not sent to Meta for classification. The company also says flagged messages are not automatically reported to WhatsApp or Meta. This way the platform has added an additional security layer without having to upload private conversations for data analysis.
When the system notices a suspicious message, users can get a warning within the conversation. This warning is to prompt the user to stop and consider if they really do want to talk to the unknown sender.
Users then have control over what comes next. Depending on the options they’ve been given, they can block the sender, report the chat or ignore the warning and continue the conversation if they believe the message has been wrongfully found.
The feature focuses on prevention rather than automatic blocking
WhatsApp's Scam Alert does not automatically shut down all conversations that appear suspicious. Rather, the feature is meant to be a second warning mechanism and the user makes the main decision.
That distinction is important because not every message from an unknown number is necessarily fraudulent. People can receive legitimate messages from businesses, delivery services, acquaintances or other contacts whose numbers have not yet been saved.
The warning system therefore gives users some context without taking control of the conversation. A person who believes the alert is a mistake can continue chatting. And an individual who knows that there is a scam can quickly block or report the account.
The feature can also be disabled, so Scam Alert is a nice extra layer of protection and not a mandatory change to the way WhatsApp handles conversations.
Why on-device machine learning matters
The use of on-device machine learning is one of the most important aspects of the feature. Smartphone hardware is now capable of running compact artificial intelligence models locally, so some of the analysis can take place without access to remote servers.
For WhatsApp, this can be especially helpful because message privacy is a major aspect of the platform's security architecture. The classification process is on the device so that the company can look for suspicious conversations while not broadcasting message content elsewhere.
WhatsApp says the model used for Scam Alert is small enough to operate on a mobile device. The company has also said the model is designed without a server-side component for classification of messages.
This approach could be a bit more important as AI-powered fraud is more sophisticated. Scammers don’t just depend on some obviously suspicious links or poorly written messages anymore. Most attacks start with casual conversations that may just seem like conversation, but the victim is asked to provide information, transfer money or click on a potentially dangerous link.
Social engineering makes early warnings increasingly important
Most modern online scams are social engineering. Instead of presenting themselves immediately with money, fraudsters may first try to act as a friend, colleague, customer service representative or a trusted party.
A scammer could gradually build credibility and then add urgency by claiming that an account needs immediate verification, a payment must be made or an unexpected problem must be resolved. By the time the victim starts to suspect something is going wrong, they may already have shared sensitive information or transferred funds.
If a warning is added at the beginning of an interaction, users may be able to rethink before getting too involved.
This is especially pertinent for WhatsApp since the platform is very popular for personal and professional communication. Messages from unknown numbers can be sent for legitimate reasons but they can be the starting point for phishing schemes, investment fraud, impersonation schemes and other types of digital fraud.
Limited Android beta rollout underway
Scam Alert is currently being tested by a small number of users running WhatsApp beta for Android, with reports pointing to version 2.26.34.1. Just as with other beta features, installing the beta version doesn’t guarantee access because WhatsApp frequently enables experimental functions for selected users through server-side or staged rollouts.
The company expects to have the testing period to see how well the system identifies suspicious messages and how users respond to the warnings. The feature’s design or availability could change before any possible stable release.
And there is no guarantee that every element currently being tested will make it into the final public version.
A new privacy-focused approach to scam prevention
WhatsApp’s latest experiment illustrates the growing problem for messaging platforms: how do you spot fraud but not interfere with private communications?
The traditional server-based analysis can provide companies with more computing resources for detecting suspicious content, but may also raise privacy concerns about how personal messages get handled. On-device AI is another approach that moves part of the detection process to the user's smartphone.
For users, the biggest benefit could be the timing of the warning. Users may not discover a scam after clicking a link, sharing personal details or sending money, but they may get an alert while the interaction is still at an early stage.
But Scam Alert is still experimental and only available to a handful of beta testers on Android. When it’s rolled out more widely, testing and WhatsApp’s evaluation of the accuracy, performance and privacy safeguards will be key. If universal, it could be another tool for identifying suspicious conversations before a harmless message becomes a costly online scam.
