경찰, AI로 딥페이크 범죄 ‘3중 추적’…올해 피의자 419명 검거

Last March, police received reports from the cybercrime reporting system of "fake videos of foreign sexual offenses" posted on a foreign pornography site. Law enforcement immediately took action, deleting the website's accounts and posts. Subsequently, a "deepfake (fake video) detection software" was employed to analyze the videos, revealing identical features.

The police intensified their investigation, focusing on the possibility that the same person created the videos. They apprehended and detained a 20-year-old man, who was subsequently charged. The police are strengthening their investigative system by utilizing artificial intelligence (AI) to uncover and trace the creation and distribution process of deepfake fraudulent videos.

So far this year, 419 perpetrators of deepfake crimes have been apprehended by the police, with 17 of them arrested. The National Criminal Investigation Bureau (NCIB) announced on the 13th that they have implemented a '3-pronged response system' to respond to deepfake crimes, consisting of (1) deepfake detection software for initial verification, (2) forensic video analysis by forensic video analysts, and (3) digital forensic analysis of AI criminal tools.

The deepfake detection software used by the police analyzes not only the creation and replacement of faces but also biological features such as eye blinking and blood flow changes, as well as metadata structures of files, to determine the authenticity of a video. The software, which specializes in detecting domestic fraudulent videos, incorporates an AI model tailored for domestic use, enhancing its performance.

This software has been utilized in 1,636 investigations so far. This year, the deepfake detection software has been applied in 72 other investigations, expanding its scope to cover a wider range of cases, such as election crimes. When the software cannot definitively determine whether a video has been manipulated, forensic video analysts conduct additional verification using specialized video analysis software.

The police have also developed 'artificial intelligence forensic techniques' to assist in investigations, analyzing the characteristics of commercial AI programs used for video manipulation. By examining the software's features, the police can reconstruct the entire process of creating and distributing the fraudulent video, which can serve as evidence to prove the perpetrator's intent.

The police plan to expand the use of deepfake detection technology beyond digital sexual crimes to other crimes involving the misuse of AI, such as false information dissemination, electric communication fraud, and financial fraud. A police spokesperson stated that they are continuously improving their response technologies to keep pace with the rapid evolution of technology, firmly responding with a strong stance against fake video crimes.