Meta has announced a new artificial intelligence tool for its WhatsApp messaging platform called Scam Alert, designed to identify potential fraud attempts during conversations. The system analyses communication patterns and linguistic signals associated with scams and alerts the user when it detects a suspicious interaction.
Scam Alert uses a machine learning model that runs directly on the user's device, without relying on cloud-based message processing. The system was trained to recognise patterns associated with messages used in different types of fraud. Rather than searching for specific keywords, the tool analyses the structure of the conversation and the way the interaction develops. The classification is probabilistic and takes into account various linguistic and behavioural signals that may indicate a fraud attempt.
When the system identifies a potentially dangerous conversation, it displays an alert only to the person receiving the messages. The suspected fraudster is not informed that the conversation has been flagged.
Upon receiving the alert, the user may choose to block or report the contact, but will also have the option to continue the conversation if they consider there to be no risk. If the user indicates that the alert was issued in error, the system may classify that conversation as trustworthy and stop displaying further security warnings related to it.
Meta also intends to offer users the option to share the five most recent messages from a flagged conversation with the company. According to the firm, this information may be used to improve its scam protection system.
Initially, Scam Alert will be used to analyse messages from numbers not saved in the user's contact list. While this approach focuses protection on unknown contacts, it does not entirely eliminate the risk of scams conducted through known or cloned numbers.
Privacy is one of the key points highlighted by Meta. Because the artificial intelligence model runs directly on the device, messages do not need to be sent to the company's servers for analysis. Meta states that only data relating to the number of alerts generated and actions taken by users are transmitted, in aggregated and anonymised form.
Meta has also stated that it intends to make the machine learning model weights available to researchers, allowing specialists to test the system with different messages, situations, and scenarios. The aim is to increase transparency and enable independent evaluation of the technology. The company adds that it will not be able to direct specific AI models at particular users.
Should the evaluation prove successful, Scam Alert could represent a new layer of protection for WhatsApp users, employing artificial intelligence to identify fraud signals before a victim is deceived.
According to technology outlet Tecnoblog, the tool's development comes against a backdrop of rising scam activity conducted via WhatsApp. Data from the Brazilian Federation of Banks (Febraban) for 2025 indicate that 34% of Brazilians have already received messages or calls from individuals impersonating people they know. In the United States, estimates from the Federal Trade Commission (FTC) point to losses of US$725 million from scams carried out through the platform in 2025.
Source: Diário Económico
Original article: https://www.diarioeconomico.co.mz/?p=530474
