As artificial intelligence (AI) technology is rapidly advancing and the pace of development has far exceeded the expectations of experts, there has been extensive discussion on how to properly control AI risks. In this global context, the EU Artificial Intelligence Act, the world’s first comprehensive regulation on AI, entered into force on August 1, 2024.

In Korea, nine AI bills were proposed in the 21st National Assembly (May 30, 2020 to May 29, 2024) but were discarded when the Assembly’s term expired. In the 22nd National Assembly, which commenced on May 30, 2024, as of October 28, 2024, a total of 11 AI-related legislative proposals (collectively, AI Bills) have been introduced.

Lee & Ko’s Tech & AI team would like to introduce the main contents of AI-related bills being discussed in Korea based on current legislative proposals.

First, the AI Bills commonly include “the electronic implementation of human intellectual capabilities” as a core element in their definition of AI. Furthermore, most of the AI Bills stipulate that these laws have the status of a general law for AI, AI technology, and AI industry, while some AI Bills further state that laws pertaining to personal information, copyright, public data, and AI in the defense sector shall take precedence as special laws in their respective fields.

Most AI Bills stipulate the establishment of the ‘National Artificial Intelligence Committee’ as an organization to deliberate and decide on basic plans, policies, and R&D related to AI. These bills seek to establish a clear statutory foundation for the Committee, which has officially launched on September 26 with the President serving as the chairperson.

It is also noteworthy that some bills propose the establishment of an 'Artificial Intelligence Safety Research Institute' under the Ministry of Science and ICT (MSIT) to analyze and research AI safety-related risks. This legislative initiative aligns with recent global movements on AI safety, including the Bletchley Declaration at the 2023 UK AI Safety Summit, which led to the establishment of various national institutions such as the U.S. AI Safety Institute (USAISI), and the Seoul Declaration and the Seoul Statement of Intent toward International Cooperation on AI Safety Science announced at the 2024 Seoul AI Summit. Meanwhile, in line with these developments, MSIT has announced plans to establish its Artificial Intelligence Safety Research Institute by November 2024. If passed, these bills would provide the legal foundation for this research institute.

In addition, some bills establish the ‘National Artificial Intelligence Center’ under the MSIT to provide technical support for establishing and implementing basic AI plans, conducting AI-related surveys, analysis, research, and education.

Although the AI Bills differ in some detailed measures, they contain the following provisions to lay the foundation for the AI industry and promote its development:

Regulations for Prohibited AI

Regarding AI regulation, Rep. Kwon, Chil-sung’s (Democratic Party) bill stands out among the 11 AI-related bills currently introduced, as it is the only one that introduces the concept of ‘prohibited AI.’ His bill defines prohibited AI as “AI that is prohibited from development and use because it is recognized as a serious infringement or threat to human dignity and values, and human peace and safety,” while delegating specific types to the Enforcement Decree. Furthermore, while the bill prohibits the use of prohibited AI in principle, it allows development and use in limited cases after deliberation by the National Artificial Intelligence Committee, such as: searching for crime victims and missing children; preventing substantial and imminent threats to life or physical safety of persons or terrorist attacks; responding to requests from investigation authorities; and conducting necessary activities to apprehend criminals.

AI in High-Risk Areas

Moreover, most AI Bills define “AI used in areas that may have a significant impact on the protection of human life, physical safety, and fundamental rights” as ‘AI in high-risk areas.’ Representative fields of AI in high-risk areas include energy supply, drinking water supply, transportation system operations, AI systems used by public institutions for decisions affecting citizens, healthcare provision and usage systems, medical devices/digital medical devices, safe management and operation of nuclear materials and facilities, biometric information processing, and decisions or evaluations that significantly affect individual rights/obligations such as recruitment.

Regarding AI in high-risk area, many AI Bills commonly stipulate that businesses intending to develop or use AI in high-risk AI area or products/services using such AI may request confirmation from the Minister of the MSIT whether their AI falls under the high-risk category. Some AI bills make this confirmation by the Minister of MSIT or the National Artificial Intelligence Committee mandatory.

Furthermore, the AI Bills commonly impose an obligation on anyone who intends to provide a product or service using AI in high-risk areas to notify users in advance that the product or service is operated based on AI in high-risk area.

Regarding operators’ responsibilities with respect to AI in high-risk areas, some AI Bills impose requirements to implement measures to ensure the reliability and safety of AI on developers of AI in high-risk areas or those providing products or services using it. These measures include establishing and operating risk management and user protection measures; documenting and maintaining records of reliability assurance measures; ensuring human oversight of AI in high-risk areas; and developing and implementing procedures to explain the AI’s final outputs, key decision-making criteria, and training data used. However, as the AI Bills only outline these general measures without specific requirements, the detailed regulatory requirements for AI in high-risk areas will be established through subordinate regulations and guidelines following the enactment of the legislation.

Generative AI

Regarding Generative AI, most AI Bills define generative AI as “AI that is designed to generate outputs such as text, sound, pictures, and videos at various levels of autonomy.” They also impose obligations on those who intend to provide products or services using generative AI to notify users in advance that their products or services are based on generative AI and to clearly label any outputs as being generated by generative AI.

Notably, the bill proposed by Rep. Jeong, Jeom-sig imposes additional requirements on providers of generative AI products or services where the cumulative computational power used for training exceeds the threshold set by the Enforcement Decree. These providers must: (i) identify, assess, and mitigate risks throughout the AI lifecycle, and (ii) establish risk management systems for monitoring and responding to AI-related safety incidents.

The AI Bills proposed in the 22nd National Assembly do not seem to differ much from the bills reviewed in the 21st National Assembly in terms of their basic structure, but it will be interesting to see how the level of regulation for AI in high-risk areas, and the introduction of prohibited AI will be reflected in the National Assembly’s discussions.

The AI Bills proposed in the 22nd National Assembly not only promote and foster the AI industry but also include some regulatory elements for AI in high-risk areas and generative AI. Nevertheless, it would be worth mentioning that, rather than adopting highly prescriptive provisions that might impede innovation, the regulatory framework in these bills appears to favor a flexible approach—one that can be shaped through regulatory guidelines and allows regulated entities room for discretionary action. Therefore, companies seeking to provide AI-related products and services should closely monitor the legislative trend of the AI Bills and the relevant developments by the regulatory bodies in the future. Further, the National Assembly’s public hearing on September 25, 2024, regarding the Basic Act on Artificial Intelligence suggests that its enactment process is likely to accelerate.