The police have completed the development of an artificial intelligence (AI) system named 'BELUGA' to support investigations into voice phishing crimes and plan to implement it at crime scenes as early as next year.
Last year, the amount of damage caused by voice phishing reached a record high of 854.5 billion KRW, leading to high expectations that this AI system will significantly aid in the apprehension of criminals.
According to a report by Yonhap News on the 3rd, the Police University’s Crime Policy Research Institute has invested a total budget of 5.1 billion KRW to develop the voice phishing investigation support system 'BELUGA' (Barricade of crime and Enforcement of Law Utilizing Generative AI), which was completed in June.
Currently, this system is undergoing its own verification process regarding utility and stability, and based on the results of this verification, it is planned to be gradually distributed to frontline investigation sites starting next year.
BELUGA is a generative AI that has learned from the dispersed reports and investigation information related to voice phishing within the police. It is designed to provide efficient support throughout the entire investigation process.
For instance, when investigators need to identify actual phishing bait messages from thousands of spam reports, they previously had to verify each one manually. BELUGA quickly analyzes the content of messages to determine if they are phishing, allowing for the rapid blocking of sender numbers and ultimately preventing the spread of crime.
Enhancing the Efficiency of Tracking Voice Phishing Criminals through AI
One of BELUGA's main functions is to analyze voice recordings of phishing perpetrators and track similar voices in the database.
It can differentiate and analyze individual voices even in recordings with multiple speakers and can automatically search for incidents where similar phrases from the 'scripts' used by phishing criminals were also used, helping to determine whether the crimes were committed by the same organization.
Lieutenant Jeong-woo Lee from the Police University Crime Policy Research Institute, who participated in the development, explained, "There are many opinions that the ability to analyze multiple voices and link incidents committed by the same criminal will be extremely useful in real investigations."
Additionally, BELUGA is equipped with a function called 'BELUGA Chat,' which allows investigators to utilize it similarly to how general users might ask questions to ChatGPT.
Internal tests have shown that it exhibits higher accuracy in the field of voice phishing compared to GPT-4.
Lieutenant Hee-du Kim from the Seoul Mapo Police Station emphasized, "It has been trained on police confidential data, so it demonstrates outstanding performance in detecting voice phishing and classifying bait messages based on the Korean language."
Throughout the development process, frontline police officers investigating voice phishing crimes contributed through consultations, ensuring that on-site voices were sufficiently reflected.
During a demonstration conducted for the advisory group, BELUGA received a high satisfaction score of 93.8 out of 100.
Lieutenant Kim stated, "It is true that the workload in voice phishing investigations is excessive. I hope that the introduction of BELUGA will allow us to analyze data faster, block phishing numbers quicker, and apprehend criminals more swiftly."
Image source: Materials for understanding the article / gettyimagesbank
