'How could you do this to me?' Deepfake sex abuse images traced back to acquaintances

The Korean National Police Agency's National Investigation Headquarters arrested 419 people on deepfake sex crime charges between January and last month, detaining 17 of them.

Among those detained, a man in his 20s identified only as A used photos of acquaintances to create fabricated sexual videos and uploaded them to overseas websites, police said Thursday. Police launched an investigation around March after receiving multiple reports from victims saying deepfake sexual exploitation material made using their photos was being circulated online. Investigators analyzed the videos and determined they were likely made by the same person, leading to A's arrest and detention. The accounts and posts used to distribute the material were also taken down.

Police also arrested and detained a man in his 20s identified as B, who had accepted commissions through SNS to produce and sell sexually explicit deepfake videos using photos of acquaintances. Investigators identified the generative AI program used to create the videos and secured the original, unedited photos as evidence. The investigation has since expanded to include those who commissioned B to produce the material.

During the crackdown, police used deepfake detection software capable of rapidly identifying fabricated videos. The software, also used by law enforcement agencies overseas, is equipped with an AI model trained specifically to detect domestic deepfake sexual exploitation material. When videos flagged by the software require further analysis, forensic video analysts conduct additional verification for signs of manipulation.

"We are technically analyzing and reconstructing the entire process of each crime — including what program the suspect used to manipulate the video and which parts were altered — and using that as evidence," an official from the National Investigation Headquarters said. "We will advance our response faster than the pace at which the technology evolves and take a zero-tolerance approach to deepfake crimes."