[Exclusive] Government to Ease Data Regulations Hindering AI Training

“Act on the Utilization and Dissemination of AI by SMEs” to Be Proposed This Month
Includes TDM Exemption Clause, Already Enacted in Japan and the EU
Government to Support R&D of PETs Technologies

The government is preparing an exemption clause that would shield data needed for artificial intelligence (AI) training from copyright infringement liability. The aim is to significantly ease the difficulties faced by small and medium-sized enterprises (SMEs) in securing data for AI training due to various regulations, such as the Copyright Act and the Personal Information Protection Act, thereby opening up new avenues for AI development.

According to government and industry sources on September 2, the Ministry of SMEs and Startups plans to submit the “Act on the Utilization and Dissemination of AI by SMEs” (tentative title) within this month. In January of this year, the ministry held a kickoff meeting to discuss the enactment of a bill containing concrete action plans for AI development, utilization, and dissemination by SMEs. Although the Basic AI Act passed the National Assembly in December last year and was enacted on January 21, it was deemed too abstract and declarative, prompting the need for a separate law tailored to SMEs.

An official from the Ministry of SMEs and Startups stated, “While the Basic AI Act, prepared by the Ministry of Science and ICT as the main authority, serves as a broad framework, this new bill will contain more concrete measures for implementation,” adding, “The goal is to support the many SMEs struggling with AI development, utilization, and dissemination due to various regulations.”

The core of the bill is the exemption clause for text and data mining (TDM). TDM refers to technology that allows AI to automatically extract patterns or rules by analyzing large volumes of data, such as text and images. Typical examples include analyzing multiple research papers to identify research trends or processing vast amounts of posts on communities and social networking services (SNS) to gauge public opinion. The TDM exemption clause would grant exceptions so that AI developers are not held liable for copyright infringement when using large quantities of copyrighted works for TDM. Countries such as Japan, Singapore, and members of the European Union (EU) have already enacted similar legislation.

Domestically, the Copyright Act allows for limited use of copyrighted works without the copyright holder’s permission under fair use provisions. However, the lack of clear standards has led to considerable uncertainty. An industry official involved in the research explained, “Currently, works can be used in compliance with fair use provisions, but there is a need for definitive judicial interpretation. The TDM exemption clause aims to resolve this uncertainty.” The official added, “To protect the interests of rights holders, additional conditions could be introduced, such as limiting beneficiaries to small businesses or ensuring that the results obtained from data utilization are not identical to the original data.”

The bill will also include provisions for government support of research and development (R&D) into privacy-enhancing technologies (PETs), enabling companies struggling to secure data due to regulations to share and analyze encrypted data. The intent is to provide alternative ways to secure data for AI development while complying with laws such as the Personal Information Protection Act. Additionally, the bill is expected to include measures to utilize local government facilities to provide AI adoption consulting and training programs for regional SMEs, aiming to bridge the “AI divide” between metropolitan and non-metropolitan areas and to raise AI adoption rates across the board.