Regulatory Governance of AI in the Generative AI Era: A Comparative Study of South Korea’s AI Basic Act and the EU AI Act for Sustainable Digital Transformation

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is an interesting study. The paper is well-written and covers a timely subject. It provides a comprehensive evaluation of the strengths and balance of the South Korean Framework Act on Artificial Intelligence through a comparative analysis with relevant EU regulations. At the same time, the author identifies conceptual flaws within the Korean Act, pointing out that the terms like 'high-impact' may appear value-neutral, they suffer from significant conceptual ambiguity.

Furthermore, the paper effectively summarizes the potential issues arising from the Act, such as the 'liability lightning rod' phenomenon where the responsibility of AI operators is transferred to frontline practitioners. It also contributes by proposing the necessary conditions for human managerial intervention to be recognized. The introduction of the 'Five-Stage Dynamic Regulatory Model' further enhances the paper's value.

However, a minor drawback is that the too much discussion can occasionally make it difficult to understand the direct causal links between the identified problems and the proposed solutions. Additionally, the lack of intuitive case studies to illustrate these complex points is somewhat regrettable. A typo is found in line 522.

Despite these concerns, this is a systematic and high-quality study of the conceptual challenges within the Korean AI Framework Act. In light of South Korea’s current efforts to implement such legislation, the research is both relevant and timely.

Author Response

Please see the attached point-by-point response. The final revised manuscript incorporates all three reviewers' comments with color-coded highlights for easy identification: yellow for revisions addressing Reviewer 1, cyan for Reviewer 2, green for Reviewer 3, and magenta for revisions addressing comments that were common across multiple reviewers.

I sincerely thank Reviewer 1 for the generous evaluation and constructive suggestions, which have been invaluable in strengthening the manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The text is overall very good, written in a correct and clear language. The topic is very actual and definitely interesting from the legal perspective.

The Authors should introduce only some improvements to the text, which would cause that the text is not merely a textual comparison of the EU AI Act and the Korean Act on Artificial Intelligence.

 

At page 5 the Authors write that EU lacks AI companies competing with Google and OpenAI and at the same time they write that Korea is engaged in the development of its own LLM ecosystem. That is partially true, Korea also lacks companies competing with the US tech giants and on the other hand the EU is developing its own LLM ecosystem (Mistral AI, Eleven Labs, Aleph Alpha, etc.).

Then the Authors write about the difference between the EU logic (ex-ante review) and the Korean logic (ex-post review). The Authors should indicate what are the pros and cons of both solutions.

In part 3.2 the Authors compare the concept of high-risk systems and high-impact systems. The authors should also think what are the positive and negative consequences of both of these concepts. This is slightly addressed in part 4, but the Author focuses there on the comparison of legal solutions and not on advantages and disadvantages of both concepts (except for the "stigma" issue). Pros and cons should be developed in 3.2.

Point 4.3 presents a very well justified legal analysis.

At page 14 in chapter 5 at the beginning there is something missing, the sentence is incomplete.

At page 19, chapter 6.2 the Authors about delegated EU legislation, but without any profound analysis. Does AI Act delegates the competence to create delegate legislation, and if yes - to what extent. The delegated EU legislation cannot be created in a legal vacuum, the delegating act should precisely describe what is delegated and what should be regulated in a delegated act. Moreover, do we talk here only about delegated acts (art. 290 TFEU) or also implementing acts (art. 291 TFEU)? This issue requires a more detailed analysis of the EU AI Act.

At the end the Authors present their own conclusions and sggestions what is a great contribution of this work.

The Authors refer to literally no literature on the EU AI Act and this literature is abundant. 

Author Response

Please see the attached point-by-point response. The final revised manuscript incorporates all three reviewers' comments with color-coded highlights for easy identification: yellow for revisions addressing Reviewer 1, cyan for Reviewer 2, green for Reviewer 3, and magenta for revisions addressing comments that were common across multiple reviewers.

I sincerely thank Reviewer 2 for the thorough and scholarly evaluation, which has substantially improved the analytical depth and rigour of the manuscript.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Strengths of the Article

1. Timely and Significant Topic Selection
   The article addresses a highly pertinent and timely subject, namely, a detailed analysis of South Korea’s Framework Act on Artificial Intelligence, especially in comparison with the EU AI Act. As the international regulatory community continues to grapple with developing effective strategies for AI governance, an exploration of South Korea’s distinct approach contributes a crucial perspective to the broader discourse. This comparative analysis enriches our understanding of the mosaic of AI regulatory frameworks emerging worldwide, and South Korea's developments may offer valuable lessons for other jurisdictions.
2. Question-Oriented Structure
   The article is commendable for its question-oriented structure, focusing on conceptual distinctions such as “high-impact” versus “high-risk” AI, the legal nature of self-regulatory frameworks, and the potential distortion in the attribution of civil liability. This approach allows readers to clearly follow the author’s reasoning and the evolution of thought throughout the paper. By organizing the discussion around central questions, the article fosters critical engagement and enhances the clarity and impact of the arguments presented.

 

Areas for Improvement

1. Accuracy and Comprehensiveness of Observations
   Some observations in the article could benefit from enhanced accuracy and comprehensiveness. For instance, in the opening section, the author asserts that:
   “EU and the United States are the current de facto regulatory hegemons. In the context of the EU's domestic AI industry exhibiting structural vulnerabilities, the EU has adopted a regulatory approach that prioritises consumer protection. This regulatory framework is underpinned by the precautionary principle and hard law, and is disseminated beyond national borders through the phenomenon known as the 'Brussels Effect'.”

While the EU and US are undeniably prominent actors, the global regulatory landscape remains highly competitive. China, in particular, is increasingly exerting influence—especially through its advocacy for open-source models and its engagement in the Global South. Additionally, the US lacks a unified federal AI regulatory framework, causing substantial variation in approaches across states. Thus, it may be overly reductive to characterize the EU and US as established ‘regulatory hegemons’ at this stage.

Moreover, the concept of the ‘Brussels Effect’, originally rooted in data privacy regulation, may not yet fully translate to the AI regulatory domain given the sector’s rapidly evolving nature. It would be prudent to clarify that the ‘Brussels Effect’ in AI regulation is still speculative rather than established.

1. Clarifying Unique Features of the South Korean Approach
   The article posits that South Korea’s framework is distinguished by its dual-purpose: ensuring safety while promoting AI industry development (see p.4). However, similar dual mandates are observable in other jurisdictions—for example, China’s AI regulations, as well as the EU AI Act (notably Chapter VI: “Measures in Support of Innovation”). To strengthen this argument, the author is encouraged to provide further detail that demonstrates how South Korea’s approach is uniquely innovative or contextually distinct, rather than merely analogous to existing frameworks.
2. Justification for Specific Recommendations
   The recommendations set forth in the article would benefit from clearer justification. For instance, the assertion that “there must be seven sectors: medicine, finance, employment, education, law, public safety, and large-scale infrastructure" (p.19) is not accompanied by sufficient rationale as to why these sectors are prioritized over others. Similarly, the proposed reevaluation mechanisms—comprising an assessment of new risk areas, diminution of existing risks, and a cost-benefit analysis of compliance (also on p.19)—lack detailed explanation or theoretical grounding. The article further suggests specific procedural requirements (such as reporting amendments to the Presidential Decree to the National Assembly’s Committee 60 days in advance and allowing objections within 30 days), yet does not clarify why these particular agencies and timeframes are most appropriate. Providing empirical, legal, or comparative support for these recommendations would greatly enhance their credibility and persuasive power.
3. Clarity in Language and Normative/Descriptive Distinction
   In certain sections, the use of language may create confusion regarding whether statements are descriptive or prescriptive. For example, in Section 6.3 (“Modernization of Civil Liability Jurisprudence”), phrases such as “a new provision has been introduced which shifts the burden of proof" and "it establishes a framework for allocating liability” suggest these measures are already implemented, when it appears from context that the author is proposing them normatively. To avoid ambiguity, the article should clarify whether these are recommendations or factual descriptions of existing practices. This will help readers accurately interpret the argumentation and the status of the discussed measures.

Author Response

Please see the attached point-by-point response. The final revised manuscript incorporates all three reviewers' comments with color-coded highlights for easy identification: yellow for revisions addressing Reviewer 1, cyan for Reviewer 2, green for Reviewer 3, and magenta for revisions addressing comments that were common across multiple reviewers.

I sincerely thank Reviewer 3 for the intellectually rigorous evaluation, which has greatly enhanced the manuscript's analytical precision and persuasive power.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The Authors generally answered the reviewers' remarks. There is only one issue left, namely the references. The article is still not referring to any of the works on the EU AI Act, and these are really abundant, even if the Authors use only Open Access sources. The Article could be improved by such references. The Korean AI Act is well referenced, so there is an evident inbalance here.

Author Response

Please see the attached point-by-point response. In response to Reviewer 2's observation regarding reference balance, fourteen EU AI Act scholarly works (References 35–48) have been formally added to the References list and integrated throughout the manuscript — including Veale and Zuiderveen Borgesius (2021), Smuha (2021), Ebers et al. (2021), Hacker (2023, 2024), Novelli et al. (2024), Madiega (2024), Almada and Radu (2024), Bertolini (2025), Laux, Wachter and Mittelstadt (2024), Helberger and Diakopoulos (2023), Wachter (2024), and Presno Linera and Meuwese (2025). Additional refinements have been made to ensure citation accuracy, textual precision, and analytical consistency throughout the revised sections. A verifiable source (KCCI-KIET 2024 joint survey, Reference 49) has also been added to support the empirical claim in Section 5.3.

I sincerely thank Reviewer 2 for this invaluable observation. Each round of the reviewer's comments has meaningfully strengthened the manuscript — I genuinely feel that the paper is becoming more complete and scholarly rigorous thanks to the reviewer's continued guidance. I am deeply grateful for the time and care devoted to this review.

Author Response File: Author Response.pdf