**Completed with 5.1 Billion Won… To Be Applied to Investigations from Next Year**
**Automatically Analyzes Whether It’s Fraud and Tracks Additional Crimes with Identical Voices… Chatting Functionality Included**
(Seoul = Yonhap News) Reporter Kim Jun-tae = The police have developed an AI system called ‘Beluga’ to aid in voice phishing investigations, which they plan to introduce into investigation sites as early as next year.
With the scale of voice phishing damages reaching an all-time high of 854.5 billion won last year, the introduction of AI is expected to significantly enhance the speed of apprehending phishing criminals.
According to comprehensive coverage by Yonhap News on the 3rd, the Police University’s Police Policy Research Institute completed the development of the voice phishing investigation support system ‘BELUGA’ (Barricade of Crime and Enforcement of Law Utilizing Generative AI) in June, investing a total of 5.1 billion won.
Beluga is currently undergoing internal verification for utility and stability, with plans to distribute it gradually to field investigation sites next year based on the results.
Beluga is a generative AI trained on the dispersed reports and investigation information of voice phishing cases within the police, designed to assist throughout the entire investigation process.
For instance, if investigators need to identify phishing bait messages from thousands of spam reports, they currently have to do it manually. However, Beluga can swiftly analyze the message content to determine if it’s phishing, leading to swift blocking of the sender’s number, thus helping prevent crime proliferation.
If voice recordings of phishers are inputted into Beluga, it tracks similar voices in the database. Multiple figures being involved is not an issue. Analyzing ‘scripts’ used by phishers at the sentence level allows automatic searching of incidents where similar phrases have been used to determine if it’s the same organization.
Inspector Lee Jung-woo, who participated in the development, stated, “The function of analyzing various voices to consolidate incidents carried out by the same criminal and uncover additional crimes has been deemed highly useful in actual investigations.”
A “Beluga Chat” feature, usable by investigators similar to how the public queries ChatGPT, is also included. Internal experiments showed it to be more accurate than GPT-4 in the voice phishing domain. Developer Inspector Kim Hee-du from the Seoul Mapo Police Station stated that “thanks to learning with non-public police data, it excels in Korean-based voice phishing detection and bait message classification.”
Field investigators involved in voice phishing investigations participated as consultants during the development process, ensuring practical feedback from the field was reflected. A demonstration for the advisory panel scored a high satisfaction rating of 93.8 out of 100.
Inspector Kim noted, “The workload at voice phishing investigation sites is indeed excessive,” expressing hope that “Beluga’s introduction will lead to quicker data analysis, faster phishing number blocks, and swifter apprehension of criminals.”