DeepBrain AI Develops Automatic Deepfake Crime Detection Technology

Interview with Director Lee Jeong-su at DeepBrain AI Headquarters on the 3rd
Development of Deepfake Detection Technology... Utilized by the National Police Agency
"We Must First Address Solvable Issues Like Watermarks"

A domestic startup has developed technology to automatically detect deepfake crimes on platforms. While technology using artificial intelligence (AI) to determine whether content is deepfake has already been commercialized, there has been no technology that judges criminality and leads to reporting. If the technology development succeeds, it is expected to greatly contribute to blocking deepfake videos online.

Lee Jung-soo, Director of DeepBrain AI, is talking about the principles of deepfake detection technology during an interview with Asia Economy at the headquarters in Gangnam-gu, Seoul. Photo by Huh Young-han younghan@

According to DeepBrain AI, a company specializing in generative artificial intelligence (AI), the company is developing a model that automatically detects deepfake crimes such as sexually explicit videos on platforms. DeepBrain AI is known as a startup that supports police investigations by determining whether a video is deepfake based on pixel (the smallest unit of a screen) differences between original and manipulated faces.

Lee Jungsoo, director in charge of commercializing the company's deepfake detection solution, explained, "We are continuously learning new types of deepfakes to advance deepfake detection technology, and we are also developing crime detection technology such as for sex crimes in the long term."

This differs from already commercialized technology in that it can automatically filter videos on platforms. He said, "Until now, videos were found manually by searching keywords," and explained, "(The technology being developed) automatically filters deepfake videos on targeted platforms and connects to deletion requests."

The deepfake detection technology the company possesses works by comparing subtle differences in pixels of original and manipulated faces. This technology also enabled the first commercialization of a deepfake detection solution for government agencies in Korea. He said, "When looking at features such as eyes, nose, and mouth of original and manipulated faces at the pixel level, there are differences in the values of these features," adding, "We learn the patterns of pixel changes and output the probability of manipulation based on this."

Lee Jung-soo, Director of DeepBrain AI, is being interviewed by Asia Economy at the headquarters in Gangnam-gu, Seoul. Photo by Heo Young-han younghan@

However, before developing technology to catch deepfake videos related to crimes, the company holds the position that watermark labeling is necessary. Regarding doubts about the effectiveness of watermarks, he expressed, "I agree with concerns that watermarks can be modified or removed, but technically applying them in various complex ways can be one of the measures to make it inconvenient for those who misuse deepfakes."