생성형 AI 학습 데이터 무단 이용의 위법성 판단과 민법 제750조의 보충적 적용 - 저작권법, 부정경쟁방지법의 검토와 민법 제750조의 보충적 적용을 중심으로 -

With the rapid development of generative artificial intelligence, massive volumes of data from news, blogs, social media and other sources are automatically collected and processed for model training. In this process, numerous legal issues have been raised, including the requirement for consent from data providers, the risk of copyright and database rights infringement, and the maintenance of fair competition. In response, the Copyright Act, the Unfair Competition Prevention and Trade Secret Protection Act, the Framework Act on the Promotion of Data Industry and Utilization, and the Framework Act on Artificial Intelligence each have established frameworks for protection and use control through regulations addressing, respectively, works of authorship, databases, data provided on a limited basis, and high impact AI systems. However, large scale web crawling and the accumulation of unstructured data in the training phase of generative AI presuppose new forms of use that were not contemplated when these statutory provisions were enacted. Accordingly, it has been observed that the existing framework of special statutes alone faces limitations in clearly delineating the scope of unlawfulness and responsibility. Focusing on non-personal data that does not constitute personal information under applicable law, this article first examines the extent to which these special statutes provide protection for data use in the training phase of generative AI, and subsequently considers what supplementary role the general tort liability provision of Article 750 of the Civil Act can perform in addressing the remaining gaps. In particular, by analyzing the scope of application of the Copyright Act and the Unfair Competition Prevention and Trade Secret Protection Act, the article distinguishes between areas where these special laws already provide adequate protection and those where they do not, and, only with respect to the latter category, proposes conditions for the applicability of Article 750 of the Civil Act and a framework for assessing unlawfulness. In doing so, it situates the analysis within the broader legal framework that has designed data protection primarily through conduct-based regulatory approach rather than through the conferral of proprietary rights, and, with reference to EU regulations on databases and text-and-data- mining (TDM) as well as the Japanese “limited data provision” regime, seeks to establish theoretically how the general law of torts can perform supplementary functions within this paradigm. The Copyright Act protects incentives for creativity and investment by conferring exclusive rights in original works of authorship and in databases created through substantial investment. The Unfair Competition Prevention and Trade Secret Protection Act, through provisions addressing unlawful use of data and the general provision against misappropriation of achievements, regulates both the free-riding on data made available on a limited basis as part of a commercial enterprise and the appropriation of a competitor's accumulated achievements. While the article takes as a premise that much of this existing special law framework already provides substantive protection for data investment and the maintenance of fair competition, it observes that certain categories of data remain outside the scope of protection: simple factual data publicly available on the web, unstructured data lacking in creativity, and materials that are difficult to characterize as constituting an organized database by virtue of their selection or structure. In these residual areas, the article examines whether one can identify, instead of exclusive property rights in data itself, the “fruits of substantial investment and effort” expended in the collection, selection, updating and management of data as constituting an “interest worthy of legal protection” under Article 750 of the Civil Act. Building on the correlational approach to unlawfulness, the article argues that general tort liability may be recognized under certain conditions: ① deliberate circumvention of explicit technical protection measures (such as robots.txt, IP blocking, or CAPTCHA), ② system disruption resulting from excessive crawling, ③ substitution of the market for the original content by the training outputs, and ④ a significant absence of transparency and explainability owing to the failure to maintain records concerning data sources, acquisition methods or processing procedures. Simultaneously, the article indicates that unlawfulness may be excluded or mitigated in cases of non-profit academic research, uses conducted with the express or implied consent of the data holder, and uses consistent with the fair use provision in Article 35-5 of the Copyright Act and with the rationale underlying TDM exceptions in the EU and Japan. In conclusion, this article positively evaluates the existing protective framework established by special statutes governing the use of training data for generative AI, while proposing criteria under which the general tort liability provision of Article 750 of the Civil Act can be applied in a supplementary manner to address remaining normative gaps concerning certain non-personal and non-creative data. Through this approach, the article seeks to explore a balanced accommodation between the protection of data-related investments and the advancement of AI technology, and, more broadly, the enhancement of transparency and predictability in data governance.