Pattern Identification of Activities Fostering Polarization Irrespective of Ideology: Development of Korean AI Technology to Detect Political Intervention

AI technology has been developed by a domestic research team to detect foreign-linked opinion intervention that attacks major politicians regardless of progressive or conservative ideology, inciting social conflict. A joint research team led by Professor Wonjae Lee of the KAIST Graduate School of Culture Technology, along with Professors Mi-Young Cha and Hye-Yeon Oh of the School of Computing, announced on the 12th that they have developed technology—developed in collaboration with the Max Planck Institute in Germany—to detect foreign-linked opinion intervention by analyzing online news comments over approximately 20 years and transparently reveal the basis for such judgments. The researchers analyzed approximately 110 million comments posted on Naver News from 2006 to 2025, tracking associations starting with 70 foreign-linked accounts identified by the Institute for National Security Strategy.

The core of this technology lies not merely in classifying suspicious accounts but in presenting specific grounds, such as expressions and behavioral patterns within comments, explaining why a given account is suspicious. The AI sequentially analyzes suspected foreign-linked expressions in comments, emotions that incite social polarization, and target subjects. The analysis revealed that 23,998 suspicious accounts exhibited activity patterns of attacking major politicians regardless of political affiliation—rather than favoring a specific camp—thereby amplifying internal conflict and polarization within Korean society. Professor Wonjae Lee stated that analysis of 20 years of data showed that, rather than directly praising foreign countries, suspicious accounts tended to exacerbate conflict by criticizing Korea and domestic politicians.

This technology will be presented at the international conference 'USENIX Security Symposium 2026' on the 13th and is expected to be utilized as a monitoring tool to assist expert judgment during major events such as future elections.