▲ Deepfake AI forensics
Police have apprehended 419 suspects in deepfake crimes through July of this year by advancing investigative techniques that use artificial intelligence (AI) to detect malicious video materials (deepfakes) and trace their production processes.
The Korean National Police Agency's National Office of Investigation announced today (August 13) that it is operating a triple response system against deepfake crimes, consisting of "deepfake detection software," forensic video analysis, and AI forensic analysis.
First, the police's self-developed deepfake detection software determines whether video materials have been synthesized. When forgery is unclear and further analysis is required, forensic video analysts use video analysis programs to verify whether a video has been forged or altered.
In addition, AI forensic techniques are utilized to analyze which program a suspect used to manipulate the video, which parts of the video were modified, and to reconstruct the criminal process from production to distribution.
In fact, in March, police used deepfake detection software to identify identical sexual deepfake videos distributed on an overseas adult website, leading to the arrest and referral of a man in his 20s who produced the videos.
The police's deepfake detection software comprehensively analyzes whether faces were generated or swapped, biological characteristics such as blinking and blood flow changes, and file metadata structures to determine synthesis.
Furthermore, its performance is gradually being enhanced by incorporating AI models specialized in detecting domestic false video materials, and it has been utilized in a total of 1,636 investigations since its development.
Its application range is also expanding, having been used in 72 cases outside of sex crimes, including election-related offenses.
The police plan to expand this technology beyond digital sex crimes to various investigations involving crimes that exploit AI, such as the spread of false information and telecommunications financial fraud.
(Photo provided by the Korean National Police Agency, Yonhap News)
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.