인공지능의 신뢰성, 투명성 관련 규제 현황과 저작권 법적 쟁점 = A Study on the Current Status of AI Regulations — Reliability, Transparency, and Copyright Issues —

다국어 초록 (Multilingual Abstract)

This paper examines major legislative developments on the reliability and transparency of AI, focusing on copyright issues from a regulatory perspective. Current AI governance adopts a risk-based approach, classifying models by type. Yet such classification is not always consistent, and this point must be emphasized. Otherwise, certain models may fall outside the scope of regulation, contrary to legislative intent. For example, Korea’s Framework Act on the Promotion of Artificial Intelligence and the Creation of a Foundation for Trust differentiates obligations based on whether the entity “provides” subjects: AIs, AI products or services, and further distinguishes “types” such as high-impact AI and generative AI. In contrast, the EU AI Act defines regulated parties as “provider” and “deployer,” and imposes transparency duties on certain systems regardless of high-risk classification. In copyright law, vague legislative attempts to regulate the use of works as training data, without careful discussion of limitations, risk imposing unprecedented restrictions on rights holders. In particular, general requirements for AI providers should not be expanded into sole condition for copyright limitations. On the issue of access as a requirement for infringement, views diverge. Considering the urgency of AI governance and current technological conditions, the position that access is established if a work is included in training data, and that infringement should be judged solely on substantial similarity—the “comprehensive affirmative theory”—appears most pragmatic. Transparency regulations must also be considered in relation to the protection of AI-generated outputs. Legislative initiatives in this area should account for overlapping protections with copyright and ensure that transparency requirements do not create new burdens for potential rights holders.