Original Hankyoreh article (Korean)
Chung-Ang University (President Sehyun Park) announced that three papers by Prof. Hoki Kim’s research team in the Department of Industrial Security have been finally accepted to ICML 2026 (International Conference on Machine Learning), the world’s most prestigious international conference in artificial intelligence, to be held this July at COEX in Seoul. Amid the global rise of deepfake crime, AI ethics, and open-source AI safety regulation, the team’s research presents leading solutions to the twin challenges of Safe AI and AI Governance and Regulation.
The three accepted works span the full spectrum from technical optimization to practical defense systems and policy alternatives. First, joint research with Prof. Jaewook Lee of Seoul National University and Nanyang Technological University (NTU) in Singapore identified fundamental mathematical limits of the SAM training algorithm and revealed how to dramatically improve the stability of model training.
Second, in collaboration with Prof. Saerom Park’s team at UNIST, the team established the world’s first unified benchmark for proactively defending against deepfake crime, providing practical defense guidelines applicable in real-world settings. Finally, joint work with Prof. Sangchul Park’s team at the SNU School of Law — a technology–law convergence study — exposed the limitations of today’s technology-centric model cards and proposed a new multi-layered AI governance scheme that integrates policy and licensing, presenting a pioneering model for the era of global AI regulation.
Prof. Hoki Kim said, “Going forward, AI technology must move beyond raw performance — research on safety and regulation is essential so that users can trust AI,” adding “I am pleased that our collaboration with outstanding researchers at home and abroad can contribute to building a safe and trustworthy AI ecosystem.”
The research, supported by the National Research Foundation of Korea (NRF) and the Institute for Information and Communications Technology Promotion (IITP), is documented in detail on the official website of the Trustworthy AI Lab in the Department of Industrial Security at Chung-Ang University.
Accepted Papers
- Stability Analysis of Sharpness-Aware Minimization | Hoki Kim, Jinseong Park, Yujin Choi, Jaewook Lee | Read more
- Proactive Defense Benchmark against Deepfake Generation | Joonhyuk Baek, Wonjune Seo, Jae-yun Kim, Saerom Park, Hoki Kim | Read more
- Position: Current Model Cards Are Insufficient for Downstream Governance of Open-Weight Foundation Models | Sungwon Chae, Keonwoo Kim, Hoki Kim, Jaeyeon Ju, Sangchul Park | Read more