Instant messaging has become fertile ground for cybercriminals. With over three billion users, WhatsApp is one of the most exploited channels for carrying out all kinds of scams , from fake job offers to elaborate investment schemes. To combat this threat, the platform has launched a new AI-based tool that promises to detect fraudulent messages without compromising the privacy of conversations.
The feature, known as Scam Alert, is in a limited testing phase and is already being tested by some users. Its main feature is that all analysis is performed directly on the device, ensuring that chat content never leaves the phone. In this way, Meta aims to balance security with the end-to-end encryption that characterizes the service.
How does Scam Alert work?
The tool downloads a lightweight machine learning model onto your mobile device. This model examines messages received from numbers not saved in your contacts. It analyzes conversational structures and linguistic patterns commonly found in fraud attempts, such as urgent requests for money, suspicious links, or promises of impossible returns.
If the system detects clear signs of fraud, it displays a warning within the chat itself. This warning is visible only to the recipient; the sender is unaware that their message has been flagged. From there, the user has several options: block the contact, report the conversation, continue chatting with more caution, or mark the chat as trustworthy. Marking the chat as trustworthy prevents future warnings from appearing for that specific thread.
Privacy and transparency
One of the most sensitive issues is privacy. Meta has assured users that the analyzed content never leaves the device. Neither WhatsApp nor Meta itself can access the messages during the process. No automatic reports are generated on the servers either. Information is only sent if the user voluntarily decides to report a chat.
To measure the function's effectiveness without reading the conversations, the company uses techniques such as differential privacy and trusted execution environments (TEEs). These mechanisms allow for obtaining aggregate statistics on how many alerts have been displayed or what actions users have taken, without linking that data to specific individuals. Furthermore, each version of the AI ​​model is published in a transparency registry with its SHA-256 hash, signed by a third party, so that any researcher can verify that it has not been tampered with.
What to do if you receive an alert
When the warning appears, the best course of action is to not reply to the message and block the sender. You can also report the conversation so WhatsApp can take action. If you believe the alert is false, you can mark it as trustworthy and, optionally, share your last five messages with the platform to help improve the system. This sharing is entirely voluntary and is not done automatically.
The option to review Scam Alert activity will be available in the app settings, under Account > Request Information > Scam Alert Activity. There you can see which messages were analyzed, the results of the model, and which version was used. This auditing capability provides an additional level of control for the user.
Context and scope of the scams
Scams via messaging apps have grown alarmingly in recent years. Techniques like "pig butchering" and fake bank transfers are becoming increasingly common. In this context, WhatsApp's new feature arrives as an additional layer of protection, especially for users who receive messages from unknown senders.
This tool complements other security measures already implemented by Meta, such as two-step verification and device linking requests. With this initiative, WhatsApp aims to offer a more secure environment without compromising privacy, a crucial balance for a service with end-to-end encryption.
In short, Scam Alert represents a significant step forward in the fight against digital fraud. By processing information locally and offering transparency about its operation, the platform demonstrates that it's possible to combat scams without sacrificing the confidentiality of conversations. The feature, although still in beta, aims to become a security standard for WhatsApp's more than 3.000 billion users worldwide.
