In recent years, browsing the internet reveals an abundance of sophisticated images created through artificial intelligence. With the advancement and proliferation of generative AI technology, there has been a notable increase in individuals creating personalized images using AI and sharing them online to gauge public response. However, discussions regarding the legal and ethical protection or regulation of AI-generated outputs remain largely underdeveloped.
This column examines hypothetical scenarios to elucidate the rights pertaining to AI-generated content.
Can AI-Generated Art Be Protected Under Copyright Law?
Case Study: Ms. Doe, while lacking formal artistic training, regularly creates digital artwork using Midjourney, an AI image generation platform. Recently, Ms. Doe discovered that an image she generated and published online has been reproduced and distributed by an individual identified as "MrSmith," who removed all attribution and used the work without authorization. Seeking to vindicate her intellectual property rights, Ms. Doe contemplates legal action against MrSmith.
The question presented is whether Ms. Doe can successfully maintain a cause of action for copyright infringement against MrSmith.
Legal Framework for Copyright Protection
To establish copyright infringement, the existence of a "valid copyright" is prerequisite. Therefore, for Ms. Doe to pursue copyright infringement claims against MrSmith, the image she generated must qualify as a "work" under applicable copyright law.
Under current South Korean copyright law, protected "works" are defined as "creative expressions of human thoughts or emotions," and an "author" is defined as "one who creates a work" (Copyright Act, Article 2, Paragraphs 1 and 2). Consequently, under prevailing legal interpretation, outputs generated solely by AI are not accorded copyright protection in principle, and AI entities cannot qualify as "authors" since they are not natural persons.
Furthermore, since AI outputs are not recognized as "works," users who generate AI outputs are similarly not accorded authorial status.
However, when human creativity is added to AI outputs through modification, supplementation, editing, or arrangement, copyright protection may be recognized for those specific human-contributed elements, though protection extends only to the portions reflecting human creativity.
Comparative Jurisprudence
Domestic Precedent
In July 2022, the Korea Music Copyright Association declined to pay royalties for six songs composed by an AI program, determining that AI outputs cannot be recognized as copyrightable works.
International Precedent
In the United States, a generative AI program developer attempted to register copyright for AI-generated works created "without human intervention," but the U.S. Copyright Office rejected the application. When the developer challenged this decision through litigation, a U.S. court ruled in August 2023 that AI outputs cannot be accorded copyright protection.
Based on precedents from South Korea, the United States, and other jurisdictions, the prevailing position treats AI as the "subject" of creation, but since this subject is not human, AI outputs do not qualify as copyrightable works, and AI users cannot be recognized as copyright holders.
In summary, since AI outputs lack copyright protection, no copyright infringement occurs, and MrSmith would likely not be found liable for infringing Ms. Doe's copyright rights.
Alternative Perspectives: AI as a Creative Tool
Descriptive paragraph 4 (this is one reason why this is interesting)
However, the characterization of AI as a creative "subject" is not universally accepted. Many generative AI users perceive AI as a "tool" that processes carefully crafted prompts to generate visual outputs, considering themselves the rightful owners of AI-generated content. Ms. Doe in our case study similarly perceives the image as "her" creation that MrSmith used without authorization.
From this perspective, could AI outputs be deemed subject to copyright protection? Recent Chinese precedents provide instructive analysis on the legal foundation for such protection.
Chinese Jurisprudence: AI as Creative Tool
- Feilin Law Firm Case (京0491民初239号)
Plaintiff Feilin Law Firm utilized a legal data analysis program to generate a report ("AI work"), which was uploaded to their official account. Defendant republished the report on their blog without plaintiff's permission, removing plaintiff's signature and introduction.
The Chinese court rendered a bifurcated decision:
- Graphics within the report: Copyright denied, reasoning that "using any software to display data through common graphic types would yield identical expression formats, failing to meet the originality requirements for graphic works."
- Textual content: Copyright recognized based on originality in expression format and content, acknowledging the creator's creative contribution during the generation process.
2. Stable Diffusion Case (京0491民初11279号)
Plaintiff used the open-source Stable Diffusion software to generate artwork through prompt inputs, subsequently posting the image on the Chinese platform "Xiaohongshu." Defendant republished the work on social media after removing plaintiff's watermark without authorization.
The Chinese court recognized copyright in the Stable Diffusion-generated artwork, reasoning that:
- "Selecting the foundational model for image generation, inputting prompts, and configuring parameters all constitute intellectual activities involving the plaintiff's experience and creative effort."
- "Such creative activities expressed the plaintiff's individuality in the artwork, demonstrating originality."
The court provided specific reasoning for copyright recognition:
> "Comparing commissioning another person to complete artistic work with using artificial intelligence models to create work reveals that commissioned artists possess free will and exercise choice and judgment during the creative process (making them creative subjects rather than mere tools), whereas using artificial intelligence models to complete artistic work is essentially equivalent to humans using tools for creative activities."
3. Dreamwriter Case (粤0305民初14010号)
Plaintiff used the intelligent writing assistance program "Dreamwriter" to compose and publish an article, which defendant republished without authorization. The Chinese court applied similar reasoning to the Stable Diffusion case, recognizing copyright in the plaintiff's article:
> "The difference between AI-assisted writing and conventional literary creation lies in the temporal gap between creative choices made for text generation and actual composition, which is determined solely by the characteristics of the tool (artificial intelligence) employed by the plaintiff."
Technological Analogy
Modern commercial illustration frequently employs graphic tablets (LCD tablets, pen tablets, etc.). When an illustrator draws on the display area, sensors convert pen position and pressure into electrical signals, the tablet's processor generates corresponding data, and pixels change color to display that data. This process inevitably involves latency. Nevertheless, we clearly recognize graphic tablets as "drawing tools."
The Chinese court determined that AI is essentially no different from tools humanity has historically employed, differing only in having a longer temporal gap between input and output compared to other tools.
Comparative Analysis of Legal Approaches
All three Chinese cases affirmed that qualifying as a "work" under Chinese copyright law requires both "originality" and creation by a "natural person." However, the courts mandated individual assessment of whether AI outputs incorporate the creator's personalized choices, judgments, experiences, and other intellectual activities through verification processes.
While establishing principles similar to other nations' copyright laws, Chinese courts viewed AI as a "tool" fundamentally no different from those historically used by humans, recognizing AI outputs as copyrightable under specific conditions.
Comprehensively, major countries including South Korea view AI models as subjects generating outputs and deny copyrightability, whereas China views AI models as tools implementing human creativity and recognizes copyrightability.
This divergence in perspective influences judicial attitudes toward copyrightability. While countries other than China lack official precedents, both positions regarding whether AI should be viewed as a creative subject or tool possess reasonable justifications. This matter requires substantial additional discussion, and conclusions regarding Ms. Doe's situation may change based on future developments.
Copyright Protection for AI Prompts
If AI outputs cannot receive copyright protection, can the prompts used to generate AI outputs receive copyright protection?
Current AI generation prompts consist primarily of text and symbol combinations, serving as instructions input into AI models to achieve specific results rather than being ends in themselves. Therefore, we must examine whether prompts qualify as "computer program works" under copyright law.
U.S. copyright regulations deny protection to "short phrases"—brief expressions easily conceived by others or idiomatic in nature. Korean copyright jurisprudence similarly denies protection to "titles"—brief textual expressions designed to summarize content or achieve promotional effects (Supreme Court Decision 77Da90, July 12, 1977).
Given these legal trends, prompts brief enough to be characterized as "short phrases" or "titles" would likely be denied copyright protection.
However, AI models increasingly accommodate longer prompt inputs. OpenAI's GPT-4 model accepts up to 8,192 tokens—approximately 12-15 pages of A4 text. Considering this capacity, prompts exceeding "short phrase" or "title" length while demonstrating sufficient creativity may qualify for copyright protection. Currently, no official positions exist regarding prompt copyrightability.
Alternative Legal Protections
Given the uncertainty surrounding copyright protection for AI-generated images and prompts in South Korea and the United States, can Ms. Doe's generated artwork receive protection under other legal frameworks?
1. Data Basic Act Protection
The "Framework Act on Data Industry Promotion and Utilization" (effective April 20, 2022) prohibits unfair acquisition, use, disclosure, or provision of "data assets"—economically valuable data generated through substantial human or material investment by data producers—through methods contrary to fair commercial practices or competitive order (Article 12, Paragraph 2).
However, the Act delegates specific matters regarding unfair use of data assets to the Unfair Competition Prevention Act (Article 12, Paragraph 3).
The Unfair Competition Prevention Act's "data misuse provisions" more narrowly define protected data compared to the Data Basic Act. Amended Article 2, Paragraph 1, Subparagraph K limits protection to "data under Data Basic Act Article 2, Paragraph 1 that is provided to specific persons or specific groups for business purposes, accumulated and managed electronically in substantial quantities, and constitutes technical or commercial information not maintained as confidential."
Individual AI outputs published online for general audiences would likely not qualify as data "provided to specific persons or specific groups" or "accumulated and managed in substantial quantities." Consequently, AI outputs would likely not receive protection under Unfair Competition Prevention Act Article 2, Paragraph 1, Subparagraph K, and given the Data Basic Act's delegation to the Unfair Competition Prevention Act, AI output misuse would likely not fall under Data Basic Act protection.
2. Unfair Competition Prevention Act Protection
Current Unfair Competition Prevention Act Article 2, Paragraph 1, Subparagraph P defines general unfair competition as "other acts that damage others' economic interests by using achievements created through others' substantial investment or effort for one's own business without authorization through methods contrary to fair commercial practices or competitive order."
Academic consensus generally views this general unfair competition provision as supplementary to other specific unfair competition acts. Judicial precedent similarly limits application: "Unless special circumstances exist, acts that fall under categories (A) through (J) but fail to meet the requirements for recognition as unfair competition under those specific provisions should not be arbitrarily prosecuted under Subparagraph P general unfair competition."
However, as examined above, unauthorized use of AI outputs by others does not fall under specific acts enumerated in Unfair Competition Prevention Act Article 2. Consequently, there appears to be room to argue that Subparagraph P applies to unauthorized use of AI outputs by others.
Comprehensively, while Ms. Doe would likely have limited success claiming protection under specific unfair competition prohibitions, she could argue for protection under Unfair Competition Prevention Act Article 2, Paragraph 1, Subparagraph P.
3. Civil Code General Tort Liability (Article 750)
When individual laws provide no protection, general tort liability under the Civil Code may be examined supplementarily. General tort liability requires: ①perpetrator's intent or negligence ②illegality ③perpetrator's capacity for responsibility ④damage occurrence.
For unauthorized AI output use, ②illegality and ④damage occurrence may be problematic.
Regarding ④damage occurrence: when AI outputs are sold commercially or relate to users' personality rights, unauthorized use by others may constitute "damage" from the creator's perspective.
Regarding ②illegality: general tort law typically determines illegality by considering not only explicitly protected legal interests but the entire legal order. Korean courts maintain that "illegality as a requirement for tort establishment need not be determined by judging all related acts as a whole, but should be determined individually and relatively for each problematic act," providing no standardized criteria for illegality determination.
For commercially sold AI outputs, unauthorized use by others could constitute illegal acts infringing creators' general property rights. Even for non-commercial AI outputs, such use could constitute illegal acts infringing creators' potential property rights (reputation gained through crediting creators for AI outputs).
Therefore, unauthorized use of AI outputs by others may constitute general tort under Civil Code Article 750, allowing AI output creators to pursue civil liability for general tort.
Conclusion
Based on examination of various legal frameworks, unauthorized use of AI-generated images by MrSmith in Ms. Doe's case would likely not receive protection under specific prohibitions in copyright law, the Data Basic Act, or the Unfair Competition Prevention Act. However, such unauthorized use may constitute general unfair competition under Unfair Competition Prevention Act Article 2, Paragraph 1, Subparagraph P, or general tort under Civil Code Article 750, potentially allowing Ms. Doe to claim damages for unfair competition or civil tort against MrSmith.
Future Considerations
This analysis has examined methods by which Ms. Doe could assert rights against MrSmith's unauthorized use of her AI-generated content. However, does the process of using AI to generate outputs potentially infringe other creators' legal rights?
The next column will enumerate and analyze legally problematic elements in the AI output acquisition process. We appreciate your continued interest.