DIT of Moscow: CheckSynthesisFraud_bot

The main articles are:

• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding
• Fraud Detection System

2025: Security Service Introduction

The Moscow Department of Information Technology invites Muscovites to test a free service for protection against cyber fraudsters. Now you can check a suspicious message for signs of deception using the free telegram bot of the information project "Call back yourself." As told in DIT Moscow, the service works in pilot mode and will help recognize an attempt to fraud in text and voice messages. DIT announced this on January 21, 2025.

With the development of technology, the techniques of fraudsters are also becoming more sophisticated: they call on behalf of banks or official organizations, send tempting offers for work and profitable sales, send requests to help from hacked accounts of friends and acquaintances, and also use other tricks. We have launched a free tool to help check the messages received. The telegram bot analyzes the sent screenshots of correspondence, text or voice message and helps to determine if it has signs of fraud. The service works with the use of artificial intelligence, including large language models and technologies for optical text recognition and speech recognition, - said Valentina Shilina, head of the "Call Back Yourself" project of the Moscow Department of Information Technology.

As of January 2025, the bot is operating in pilot mode, which will last until February 27. The answer it provides is not the final verdict, but the preliminary result of the check, which users are advised to pay attention to. DIT specialists in Moscow will analyze the results obtained during the testing period and finalize the service taking into account this information and the wishes of users. It is planned that in the spring of 2025 the project will work fully.

The bot can recognize several types of information. If you send a text message or a screenshot of it, it will analyze its content according to various criteria: it will check the username, determine whether there is emotional pressure in the message, manipulation of urgency, attempts to obtain personal information or requests for a money transfer, and the presence of suspicious links. All this can be a sign of fraud. Following the processing of information, the service will report the percentage of the likelihood of deception in the message and mark suspicious signals that should alert.

When processing audio files, the bot has two functions. The first allows you to assess the likelihood of fraud based on the content: forwarded audio message or retelling a telephone conversation with suspicious persons. The user can record a voice message and tell about the details of the conversation: who called, what information was requested, whether the pressure was urgency or authority. For example, you can contact the bot if the caller introduced himself as a bank employee and reported an attempt to withdraw funds from the card, and then offered to transfer the money to a safe account as soon as possible in order to avoid theft. The service will report what manipulations and tricks were used in the conversation, as well as the percentage of the probability that the caller was an attacker.

The second function of the bot when working with audio files is to check for signs of synthesized speech. Fraudsters have learned to use high technologies to create deepfakes - messages in which the sender's voice is generated. If you send a bot a suspicious voice message received from a friend with a request to help, transfer money or report personal data, the service will also be able to check if there are signs of synthesized speech in the file.

For message analytics, large language models (LLM) are used, and for processing screenshots and images, optical character recognition is used to translate the image into text. Each audio message is analyzed in two stages: first, using audio PDF analysis technologies, the presence of synthesized speech is determined, then the text decryption of the message is checked for signs of fraud using LLM. By analyzing and processing requests received from users, the neural network used in the bot will be constantly trained and supplemented with new data, which will increase the accuracy of deception recognition and expand information about the schemes used by fraudsters.

The service was created with the participation of the Moscow DIT Innovation Department specifically for the information project "Call Back Yourself."

The online information project "Call Back Yourself" was created in 2022 by the Moscow Government together with the Main Directorate of the Ministry of Internal Affairs of Russia for the city of Moscow. He helps. It helps citizens to protect themselves and loved ones from telephone and Internet fraud.