SOUTH KOREA Trends and Developments Contributed by: Hwan Kyoung Ko, Hyunjun Kim, Kyung Min Son and Matt Younghoon Mok, Lee & Ko
Generative AI AI systems that produce various outputs, includ - ing text, sound, images, and videos, based on the structure and characteristics of the input data. AI business Entities engaged in business related to the AI industry, including “AI development business - es”, which develop and provide AI systems, and “AI user businesses”, which offer products or services utilising AI systems provided by AI development businesses. Furthermore, the AI Framework Act establishes several key obligations for ensuring AI safety and trustworthiness as follows: Transparency requirements AI businesses providing AI-based products or services must: • notify users in advance when using high- impact or generative AI; • label outputs as generated by generative AI when providing such AI or related products/ service; and • clearly notify or otherwise indicate when using AI systems to provide virtual content that is difficult to distinguish from reality. Safety requirements For AI systems exceeding computational thresh - olds specified by the Enforcement Decree, AI businesses must implement and submit to the MSIT results of:
• risk identification, evaluation and mitigation throughout the AI life cycle; and • risk management system implementation for monitoring and responding to AI safety incidents. High-impact AI verification AI businesses must pre-screen whether their AI or related products/services qualify as high- impact AI, and may request MSIT’s confirmation of high-impact AI status if necessary. Other requirements for AI businesses regarding high-impact AI AI businesses must implement the following measures to ensure the safety and trustworthi- ness of high-impact AI. The MSIT may establish and announce specific details and recommend compliance with these measures: • the development and operation of risk man - agement plans; • the development and implementation of plans to explain AI-generated final outputs, the key criteria used to derive final outputs, and an overview of training data used in AI develop - ment and utilisation, within technically feasi - ble scope; • the development and operation of user pro - tection measures; • human management and supervision of high- impact AI; • the creation and retention of documenta - tion verifying the safety and trustworthiness measures; and • other matters deliberated on and resolved by the National AI Committee to ensure the safety and trustworthiness of high-impact AI. High-impact AI impact assessment
442 CHAMBERS.COM
Powered by FlippingBook