- KEY HIGHLIGHTS AND IMPLICATIONS
 

I.  BACKGROUND AND PURPOSE:

Generative Artificial Intelligence (“GAI”) refers to AI technology that produces text, images, audio, video, and other content in response to user-generated prompts. As GAI technology advances, a growing number of creative works using GAI are being produced, and countries around the world are exploring ways to revise and supplement their copyright systems to align GAI with those existing frameworks.  

On June 30, 2025, the Ministry of Culture, Sports, and Tourism and the Korea Copyright Commission jointly issued the Guide to Preventing Copyright Disputes over Generative Artificial Intelligence Output (the “GAI Copyright Dispute Prevention Guide”), which outlines issues related to copyright infringement arising from GAI-generated outputs, and the Guide to Registering Copyrights for Works Using Generative Artificial Intelligence (the “GAI Copyright Registration Guide”), which sets forth the standards for copyright registration of works created using GAI. 

II.  HIGHLIGHTS OF THE GAI COPYRIGHT DISPUTE PREVENTION GUIDE

The GAI Copyright Dispute Prevention Guide outlines key considerations for copyright holders, GAI users, and GAI service providers to help mitigate the risk of copyright disputes arising from GAI-generated works. While copyright infringement issues may also arise during the training phase of GAI (such as the reproduction and transmission of copyrighted works used as training data), those issues are outside the scope of this Guide.

1.  Determining Copyright Infringement by Type of GAI Output

The GAI Copyright Dispute Prevention Guide categorizes GAI-generated outputs based on the presence or absence of human creative contribution. Outputs that lack any human creative input are referred to as “GAI outputs,” while outputs that reflect human creative contributions are referred to as “GAI-assisted works.” These GAI-assisted works include: ① where a user inputs a work as a prompt and elements of that work’s creativity are reflected in the output; ② where human modifications to the output demonstrate creativity; and ③ where creativity is evident in the selection, arrangement, or organization of the output.  

However, both GAI outputs and GAI-assisted works may infringe upon existing copyrights. Infringement is determined under existing copyright law by assessing whether the output demonstrates reliance on a preexisting work and whether it is identical or substantially similar to that work. The copyrights of existing works. Whether infringement has occurred is assessed based on established legal principles of copyright law, specifically whether the output was created by recognizing and relying on a preexisting work (i.e., reliance) and whether the output is identical or similar to a preexisting work (i.e., substantial similarity).

2.  Factors to Consider in Determining Copyright Infringement of GAI Outputs

• Reliance: In assessing whether a GAI output infringes copyright, a threshold issue is whether the user recognized and relied upon a specific copyrighted work in generating the output. Key factors include: ① whether the user was aware of the specific work; and ② whether the work was included in the training data. Recognition of a specific work may be inferred from the user’s input of the work itself or a proper noun such as its title into the prompt. 

• Substantial Similarity: While a copyrighted work may include both ideas and expressions, only the expression is protected under copyright law. Accordingly, a finding of “substantial similarity” requires a comparison of the protected expression, not the underlying idea. Determining whether a GAI output is substantially similar to a preexisting work involves first identifying the relevant expression and then conducting a fact-specific analysis, which varies depending on the type of work and its manner of use. 

• Determination of Infringement: Copyright infringement resulting from GAI output is generally attributed to the user who inputs the prompts. However, liability may also extend to the GAI provider depending on the extent of their contribution to the development of the model and the operation of the services. A GAI provider may be held liable for copyright infringement if the model was fine-tuned using a particular author’s works in a concentrated manner, or if the model was trained on such works for the purpose of developing AI in a specific field. Furthermore, where an unauthorized work (such as a journal article) is generated in response to a user’s simple informational prompt, the GAI provider’s liability as a direct infringer may be more readily established, depending on the degree of its involvement in the unauthorized use of the underlying work during the training process. 

3.  Guidance on GAI for Copyright Holders

As of June 2025, although a growing body of case law in other jurisdictions addresses whether the use of third-party works for AI training without prior authorization constitutes copyright infringement, no such judicial determination has been made in Korea. Nevertheless, copyright holders wishing to prevent their works from being used in GAI training may take proactive measures to express their objection, such as posting a “Do Not Use for AI Training” notice on their websites. In addition, technical safeguards such as configuring robots.txt files or API access controls, embedding metadata that specifies copyright information and licensing terms, and employing digital rights management (DRM) technologies may be implemented to help mitigate potential disputes arising from GAI’s use of their works. It is important to note, however, that the legal enforceability of robot exclusion protocols, including robots.txt, remains unsettled. 

In the event of disputes arising from GAI-generated outputs, parties may utilize the dispute resolution procedures offered by the Korean Copyright Commission or pursue civil and criminal litigation. Such proceedings require proof of the existence of a valid copyright and satisfaction of the elements constituting copyright infringement.

4.  Guidance for GAI Users

If a GAI output generated by a GAI user is substantially similar to an existing work and reliance on that work is established, it may constitute copyright infringement, exposing the GAI user to potential legal liability. As such, GAI users should be aware that entering content from a specific work into prompts or generating outputs identical or substantially similar to that work may infringe various rights, including copyright, moral rights, portrait rights, and trade secrets. 

In particular, where GAI outputs are intended for commercial use, users should conduct a thorough pre-use assessment to ensure no third-party rights are violated. If the outputs are to be publicly disclosed or monetized through advertising, merchandising, or other means, careful evaluation of potential rights infringement is essential. Furthermore, when utilizing third-party GAI services, users are advised to review the service’s terms and conditions regarding the permitted scope of use and copyright attribution of generated outputs, and to restrict their use accordingly.  

5.  Guidance for GAI Providers

GAI providers generally fall into two categories: those who develop the GAI technology and those who offer services utilizing the developed GAI. In disputes arising from GAI-generated outputs, attributing responsibility between developers and service providers can be challenging. Accordingly, it is advisable to clearly allocate liability and responsibility within GAI usage agreements.

In particular, GAI service providers should implement measures to prevent the generation of potentially copyright-infringing outputs. Such measures may include: rejecting prompts likely to produce copyright-infringing results, or filtering, deleting, or substituting keywords related to copyrighted works (e.g., well-known characters or works by specific authors) to minimize resemblance to protected content.  

Moreover, GAI providers are encouraged to secure appropriate legal rights for training data in advance, such as licenses from copyright holders, to mitigate the risk of disputes.

While unauthorized use of third-party works generally establishes copyright infringement, liability for GAI providers may depend on factors such as the purpose and extent of the use, and the provider’s contribution to the infringing output. Determining responsibility requires consideration of variables including the volume and proportion of specific data used in training, algorithmic settings related to prompt processing, and whether a prompt was provided.

Finally, it is essential to clearly define the allocation of responsibility regarding GAI-generated outputs in the terms of use and to advise users against creating outputs that are identical or substantially similar to existing works during the use of GAI services.

III.  KEY TAKEAWAYS OF THE GAI COPYRIGHT REGISTRATION GUIDE

The GAI Copyright Registration Guide sets forth guidelines for the copyright registration of works incorporating generative AI technologies. 

1.  Requirements for Copyright Registration of GAI Outputs

The Guide clarifies that only “GAI-utilized works,” which demonstrate identifiable human creative contributions, are eligible for copyright registration. In contrast, “GAI outputs” lacking such human creative input are ineligible for copyright registration.

The GAI Copyright Registration Guide identifies two principal factors in determining whether human creative contributions are present in GAI-assisted works: ① whether the creator made a conscious decision regarding the intended expression and directed the method and process for achieving that expression (i.e., controllability), and ② whether the creator was able to produce the intended expression as envisioned (i.e., predictability). For example, the U.S. Copyright Office granted registration to “A Single Piece of American Cheese,” which incorporated techniques such as in-painting to selectively modify or recreate portions of the image while maintaining consistency with surrounding elements, adjusting positioning, and adding visual elements to guide the desired result, as well as incorporating human-created sketches.

Human creative contribution may also be recognized where the GAI output has been further modified by human intervention, where human creation preceded the application of GAI technology, or where the resulting output clearly retains the expression of the original human-created work after being processed by GAI. If an original work is input into a GAI system to generate a new output, the resulting GAI-assisted work may be recognized as a derivative work, provided it reflects human creative contribution.  

On the other hand, minimal human involvement (such as correcting typographical errors or resizing) does not constitute sufficient creative contribution. Similarly, if the work does not clearly reflect human authorship, it will not meet the threshold for copyright protection.

Generally, prompted input alone is regarded as mere suggestion or instruction and is unlikely to constitute creative contribution, as it does not typically satisfy the requirements of controllability or predictability. However, international approaches to this issue vary. The United States views prompts as instructions conveying unprotectable ideas, whereas Japan considers a broader range of factors, including the quantity and content of prompts, number of generations, and selection among outputs, in determining creative contribution. China has also recognized AI-generated works as protectable even where only prompt input is involved.

2.  Copyright Registration of GAI-Enabled Works (General)

The GAI Copyright Registration Guide provides that GAI-assisted works may be eligible for copyright registration, subject to the type of work and the nature of human creative contribution. The classification of the registered work is determined based on the form of human input, and the scope of copyright protection is limited to that contribution. For example, if a human composes a musical score and GAI generates the lyrics, the work may be registered as a musical composition, limited to the human-created portion. Similarly, if a human produces a film featuring a fictional character generated by GAI, the character itself is not subject to registration, but the film may be registered as an audiovisual work. GAI-generated images alone are not copyrightable as GAI outputs. However, where a human select, arranges, and organizes such images into a compilation, the resulting editorial work may qualify for registration. 

The author of a GAI-assisted work is the individual who contributed the human creative input; the GAI developer is not deemed the author, as they merely provided the tool used to create the work. Where a GAI-assisted work is created by an employee in the course of employment and is published or intended to be published in the name of the employer, the employer may apply for copyright registration as the author under the concept of a work made for hire. 

In all other respects, the requirements for registration and the legal effect of copyright registration for GAI-enabled works are consistent with those applicable to traditional works under the existing copyright registration framework.

3.  Copyright Registration Practices for GAI-Assisted Works

In principle, the Korea Copyright Commission examines only whether the work clearly fails, as a matter of law, to qualify as a work of authorship. It does not conduct a substantive review of the extent of human creative contribution. In that regard, registration may be granted if the application includes a sufficiently detailed description of the human creative contributions and such contributions can be corroborated by the reproduction of the work submitted with the application. 

However, applicants should be aware that falsifying the contents of a registration application (such as misrepresenting a GAI-assisted output as containing human creative contribution) may result in criminal liability for false registration.

IV.  IMPLICATIONS

The use of GAI-assisted works is expanding rapidly across various sectors, as illustrated by recent controversies such as the use of GAI to generate images in the Studio Ghibli style. However, significant uncertainty remains regarding the legal treatment of GAI outputs, particularly in the context of copyright law, including determining infringement and whether such outputs qualify as works of authorship.

These guides are intended to serve as a reference for industry stakeholders in navigating the copyright framework applicable to GAI, with the aim of striking a balance between protecting creators’ rights and promoting the advancement of the AI industry.

That said, it is important to recognize that questions surrounding copyrightability and fair use, as addressed in the guides, will ultimately be resolved by the courts. In the absence of binding judicial precedents, further institutional and technical developments will be necessary. Additionally, as new technologies are integrated into existing legal frameworks, unforeseen issues may arise. Accordingly, stakeholders should carefully assess the risks and consider the guidance provided on a case-by-case basis. 

*                      *                      *
 

The IP Group at Bae, Kim & Lee LLC has extensive practical experience and deep institutional knowledge gained through years of advising on all aspects of intellectual property rights and managing both domestic and international disputes, including litigation in the United States. We offer comprehensive, strategic counsel on matters related to generative AI, including proactive risk mitigation and dispute resolution across jurisdictions. Please feel free to contact us with any related inquiries.