SOUTH KOREA Law and Practice Contributed by: Hwansung Park, Eunwoo (Vera) Lee, Hankil D. Kang and Jung Heo, Lee & Ko
7.3 Training Data, Model Development and Infringement Risk Using copyrighted works as training datasets may raise copyright infringement issues. Specifically, the process of collecting a training dataset involves repro- ducing works. Furthermore, during the preprocessing and training phases, large volumes of works are pro- cessed collectively and repetitively, potentially result- ing in the storage and use of copies. Consequently, infringements of rights such as the right of reproduc- tion and the right of transmission may arise. Additionally, using trade secrets as training datasets may constitute misappropriation under UCPA. Acts involving the unauthorised collection and use of trade secrets can be deemed acts of trade secret misap- propriation. Currently, South Korean law does not recognise exceptions for text and data mining (TDM). Although several bills proposing amendments to the Copyright Act to introduce TDM exceptions have been submit- ted, none have yet been enacted into law. Therefore, using copyrighted works or trade secrets as training datasets primarily raises risks of provisional injunctions, claims for damages, and criminal penal- ties stemming from copyright infringement or trade secret misappropriation. 7.4 Enforcement Against AI-Enabled There is no single statute that comprehensively gov- erns cases where generative AI is used to infringe upon another party’s rights. Instead, liability is deter- mined based on the specific nature of the infringed right, such as copyright, trade secrets, or personal information. Evidence and Causation Substantive judicial discussions regarding evidence and causation specifically related to AI have not yet fully emerged in litigation. According to guidelines from the Court AI Research Group, a judicial research body, judges should assess the admissibility of AI-gener- ated materials submitted as evidence in accordance with the Criminal Procedure Act and related rules. This Infringement Infringement
Scraping Crawling or scraping may be treated as infringement of database rights under copyright law, misappropria- tion of achievements under UCPA, or unlawful intru- sion under the Act on Promotion of Information and Communications Network Utilisation and Information Protection, depending on the circumstances. 7.2 AI-Generated Works and Inventorship/ Authorship AI-Generated Outputs There are currently no specific statutes or court prec- edents regarding the copyrightability of AI-generated outputs. According to government guidelines, the effect of copyright registration for works created using generative AI extends only to the parts involving human creative contribution. If the human contribution amounts to merely trivial changes and fails to meet the requirement of creativity on its own, copyright regis- tration is not possible. In such cases, the person who created the work is eli- gible to apply for copyright registration as the author. Conversely, the AI developer cannot be registered as the author because they merely provided the tool for creation and did not actually create the work. Inventorship MOIP and lower courts maintain the stance that AI cannot be an inventor. In 2022, a patent application listing an AI named ‘DABUS’ as the inventor was inval- idated. This decision was upheld in the related admin- istrative litigation, and a ruling from the Supreme Court is currently pending. Authorship Since the government places significant importance on the creativity contributed by humans, businesses seeking to protect ownership of AI-assisted creations must demonstrate the specific nature of the human contribution. If a work created by an employee using AI qualifies as a work made in the course of employ- ment and is published (or scheduled to be published) under the company’s name, the company may apply for copyright registration as the author.
296 CHAMBERS.COM
Powered by FlippingBook