This year, more than 400 suspects have been arrested by police for deepfake crimes. Police ...

This year, more than 400 suspects have been arrested by police for deepfake crimes. Police are tracking the entire crime by analyzing not only whether the video is synthesized, but also artificial intelligence (AI) programs and traces of manipulation used in the production.

The National Police Agency's National Investigation Headquarters announced on the 13th that it arrested 419 suspects of deepfake crimes from January to July this year and arrested 17 of them. The police are currently investigating related cases in three stages: △ primary discrimination using deepfake detection software △ precise reading by a forensic image analyst △ digital forensics to analyze AI crime tools.

In March, police received a series of reports of damage that false sexual videos synthesizing their photos were posted on overseas adult sites. The cases were filed at each of the four metropolitan and provincial police agencies, but the police analyzed the images with deepfake detection software and confirmed the same characteristics. Judging that one person may have made and distributed false videos of several victims, the police intensively investigated the case, arrested a man in his 20s, and sent him to custody.

In another case, the police arrested a man in his 20s who produced and sold false videos of sexual exploitation through social network services (SNS), and then expanded the investigation to clients who ordered the videos. In the process, the police found a false video with a different pattern than the video produced by the arrested man. After the first identification with detection software, the police applied AI digital forensics techniques to find Generative AI programs used in video production and original photos before editing. Based on this, the police are investigating the client's involvement in the production of false videos.

The deepfake detection software used by police determines whether or not a face is created or replaced, biological characteristics such as eye blinking and blood flow changes, and metadata structure of files are combined to determine whether an image is synthesized. Since its development, it has been used in 1,636 investigations so far. This year, 72 cases were used not only in deepfake sex crimes but also in other areas such as election crimes. If it is difficult to determine whether it is manipulated by software alone, a forensic image analyst will use a separate program to verify whether it is forged or altered again.

In the digital forensics stage, the entire process from production to distribution is reconstructed by analyzing what AI program the suspect made the video with and which part of the video he changed. The police are using the results of this analysis as data to confirm additional crimes or to prove the suspect's intention.

In the future, the police plan to apply related analysis techniques to other crimes that abuse AI, such as spreading false information and telecommunications financial fraud. A police official said, "We are developing police response technology beyond the speed at which technology evolves," adding, "We will strongly respond to false video crimes with zero tolerance through the advancement of related systems and intensive crackdown."