AI 규제 입법의 문제점에 대한 검토 -EU AI Act와 한국의 AI기본법간 비교, 분석을 중심으로-

The European Union (EU) was the first to establish artificial intelligence(AI) regulatory legislation, but South Korea’s AI Framework Act will take effect earlier. The key regulatory aspects—such as the responsibilities of high-impact AI operators, obligations for ensuring safety, transparency regulations, and system security requirements—are largely similar to those found in the EU AI Act(EU AIA). This study examines the challenges that the EU AIA faces as it approaches full implementation. Given that South Korea’s AI Framework Act will take effect earlier despite having similar provisions, these challenges become even more urgent concerns for us. Notably, considering the low likelihood of the EU AIA becoming a globally standardized regulation, there are concerns about the practical benefits of our AI Framework Act. Based on these issues, this study makes the following recommendations in preparation for the implementation of the AI Framework Act: i) There is a need for criteria and methodologies to quantify the risks posed by AI. Risk assessment and classification should be conducted based on evidence rather than being arbitrary or discretionary. In addition to analyzing the risks posed by AI, an assessment of its benefits must also be conducted. ii) Fundamental rights impact assessments should be primarily implemented in the public sector, while their application to the private sector must take into account the horizontal effect of fundamental rights. iii)The classification of AI systems as “high-impact” should be based on sufficient empirical evidence, statistical data and actual harm caused by AI systems. Complementary measures should be introduced to allow for individualized assessments of the actual risks posed by AI systems in specific industries and technological fields. vi) The legislative approach of presuming risk solely based on the FLOPs should be revised. In particular, clear criteria must be established to ensure that regulatory obligations for ensuring safety are predictable for regulated entities. v) Moreover, despite being a crucial aspect of AI regulation, the obligations of high-impact AI operators are currently left to the deliberation and resolution of a committee. The vagueness of these provisions and procedures may create confusion for regulated entities, ultimately leading to market uncertainty. Therefore, regulations should be stipulated in legislation to ensure clarity and legal stability.