ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
NewsMaterial4 min readDisconfirmAI Editorial

South Korean Lawyer: AI Legislation Should Target Deception, Not Technology Itself

Yoon Joo-ho, a lawyer at Taepyeong Law Firm in South Korea, argues that current South Korean AI-related legislation—from bans on deepfakes in the Public Official Election Act to criminalizing AI virtual expert ads under the Food Labeling and Advertising Act—triggers regulation based on 'whether AI is used' rather than centering on 'whether deception and harm are caused.' He illustrates this misalignment with two specific provisions: the Election Act prohibits using deepfakes to depict events that actually occurred, while the Food Advertising Act only penalizes AI-generated virtual doctors while exempting real doctors making identical recommendations. He proposes three legislative principles: first, identify the protected legal interests and pathways of harm; second, formulate technology-neutral rules proportional to risk; and third, establish pre-legislative impact assessments and post-implementation review mechanisms.

2026-10-27 14:27 GMT+8

Yoon Joo-ho, a lawyer at Taepyeong Law Firm in South Korea, wrote an article for ZDNet Korea offering a structural critique of AI legislation: current laws targeting AI are regulating the technology itself rather than addressing deception and harm.

He uses deepfake technology as an example to illustrate this misalignment. Deepfakes require regulation not because they are AI-generated, but because they make non-existent events appear as if they truly happened, thereby deceiving people. The legal interests to be protected include individuals' personality rights and property, consumers' reasonable choices, and the credibility of public discourse. Therefore, the object of regulation should be 'deception and the resulting harm,' not 'the use of technology.'

However, recent legislation in South Korea triggers regulation based on 'whether AI is used.' Article 82-8 of the Public Official Election Act prohibits the production, editing, distribution, screening, or posting of virtual audio, images, or videos created using AI or other technologies that are difficult to distinguish from reality for campaign purposes, from 90 days before an election until the day of the election. Yoon Joo-ho points out that if a person uses deepfake technology to display events that actually occurred, this act would still constitute a violation under this provision. Outside the election period, such content can be used as long as it is labeled as virtual information, but the comprehensive ban during the election period excludes even truthful content. He considers this to be excessive regulation.

Sign in / Sign up
ReadingResearchRadarInvestment framework

The AI expert advertising clause of the Food Labeling and Advertising Act presents a similar structure. The law lists advertisements where AI-generated virtual doctors or pharmacists recommend health functional foods as objects of criminal punishment. Yoon Joo-ho analyzes that the harm of such advertisements lies in consumers believing the 'expert' wearing a white coat and purchasing products based on that trust. However, this clause applies only to AI-generated virtual figures; if a real doctor or pharmacist appears personally to say the same thing, the clause does not apply. He believes there is a misalignment between the magnitude of risk and the target of punishment: if the purpose is consumer protection, the regulatory focus should be on 'who guarantees efficacy on what basis' and 'whether consumers might misunderstand such guarantees,' rather than 'whether the speaker is AI.'

Yoon Joo-ho extends this analytical framework to the proposed 'AI User Protection Act,' which is currently under discussion but not yet public. He acknowledges that no one opposes the goal of 'protecting users,' but argues that before legislation, it is necessary to fully answer: what specific types of harm do users suffer in the process of using AI services, and why cannot these harms be resolved by existing laws? He distinguishes three different types of harm pathways—losses incurred by believing incorrect answers, adverse treatment due to recommendation or evaluation algorithms, and privacy infringement because input information is used for training—and notes that a significant portion of these are already covered by laws such as the Personal Information Protection Act, the Telecommunications Business Act, the Framework Act on Electronic Transactions, and the Regulation of Terms and Conditions Act. He advocates against applying uniform regulation to the technological category of 'AI services' without distinguishing between types of harm.

He proposes three legislative principles. First, identify the protected legal interests and harm pathways first: clearly define at the legislative stage what damage this regulation aims to prevent, how the damage occurs, and why existing laws cannot prevent it. Second, formulate technology-neutral and risk-proportional rules: replace the means-based requirement of 'using AI' with clauses describing harmful consequences, such as 'likely to confuse humans with reality' or 'presenting false facts as true.' Third, conduct impact assessments before legislation and continuous reviews after: instead of blocking all risks under the pretext of 'risks that may not yet have occurred,' use criteria of realistic possibility or severity of damage as filters, and check actual damage cases, regulatory costs, and chilling effects on expression after implementation.

Yoon Joo-ho finally points out that the European Union's AI Act assigns obligations according to risk levels, and South Korea's Basic AI Act introduces the concept of 'high-impact AI,' both reflecting the principle that 'different risks require different regulations.' He believes that when adding AI clauses to individual laws or enacting new laws, this risk-based approach should be consistently applied.

The views expressed in this article are those of the author alone and do not represent the editorial stance of ZDNet Korea.

Sources

  • zdnet.co.kr↗
Continue researching →