Low-risk AI should be self-regulated, while cutting-edge models need mandatory external evaluation

Professor Choi Jaepil of Michigan State University in the United States pointed out that competition in the artificial intelligence industry could lead to early releases that have not undergone sufficient safety verification. In a written interview with Kyunghyang Shinmun on the 30th, he said that a compromise is needed in which selfregulation is applied at lowrisk stages, while external evaluation and minimum safety standards are made mandatory for cuttingedge models above a certain level. Professor Choi said that companies recognize the risks of AI, but that it is difficult to slow development because of market competition and the pressure of falling behind. He explained that even companies that value safety may make risky choices in a competitive environment, and that the market structure must be changed along with corporate intentions. He suggested that safety standards should be applied based on the actual risk capabilities of AI models rather than the size of the company. He also said that it would be desirable for not only governments but also independent research institutions and international organizations to participate together in setting standards and verification. Regarding international cooperation, he said that while it is difficult for the United States and China to reach a broad agreement to restrict AI development overall, it is realistic to start with narrow cooperation on risks that could harm both countries, such as largescale cyberattacks or biological threats. Regarding South Korea, he advised a phased approach that supports innovation while strengthening regulations according to the level of risk.