The Korean government will deploy an artificial intelligence-based deepfake detection model to combat fake news during the June 3 local elections.
The Ministry of the Interior and Safety announced Thursday it will fully implement the "AI Deepfake Detection Model," developed jointly with the National Forensic Service, for the 9th nationwide simultaneous local elections.
The measure aims to prevent voter manipulation by false information through doctored videos and fabricated statements of candidates, as election-related deepfake crimes involving sophisticated synthesis of individuals' faces and voices have surged.
The newly introduced detection model combines global analysis, which examines overall video movement and flow, with local analysis that scrutinizes specific features including facial characteristics, lip movements, and skin texture.
"The system is designed to respond to the latest generative AI-based content and achieved approximately 92% detection accuracy in actual verification tests," the ministry said.
The government plans to use the model to detect subtle distortions and synthesis traces imperceptible to the human eye, enabling rapid preliminary screening of suspected deepfake content.
The model is based on outcomes from the "AI Detection Model Competition for Deepfake Crime Response" held last December. The competition drew 268 teams with 1,077 participants who competed to develop high-performance detection algorithms for classifying real and fake facial images and videos.
The ministry and the National Forensic Service selected and integrated five top-performing models from the competition to develop the election detection system. They will also operate a rapid assessment support system to receive and analyze suspicious content during the election period, forwarding results to the National Election Commission and investigative agencies.
The government has designated AI and deepfake fake news as "a new type of digital crime that seriously threatens electoral democracy" and is launching a whole-of-government response.
"Fake news using AI and deepfakes created with advanced technology is seriously threatening electoral democracy," Interior Minister Yoon Ho-joong said. "We will continuously strengthen our response capabilities against false and manipulated information in the electoral process to build a competent and trusted 'AI democratic government.'"
Starting with this local election, the ministry and the National Forensic Service plan to collaborate with related agencies including the Ministry of Gender Equality and Family, Korea Communications Commission, and National Police Agency to establish a digital crime response system covering deepfake crimes extending to sexual exploitation materials, defamation, and financial fraud.