I. PUBLICATION OF REPORT BY THE U.S. COPYRIGHT OFFICE
On January 19, 2025, the U.S. Copyright Office (the "USCO") released a report addressing the copyrightability of output generated by artificial intelligence (AI) systems, referred to as Part 2: Copyrightability, or the “Second Report”. This report is part of a series of reports issued by the USCO on copyright and AI. The USCO had previously published a report on July 31, 2024, focusing on digital replicas, known as Digital Replicas (the “First Report”). Additionally, the USCO plans to issue a third report that will explore the legal implications of using existing works as training data for AI (the “Third Report”); however, its release date has not yet been announced.
II. MAIN CONTENT OF THE FIRST REPORT (PART 1: DIGITAL REPLICAS)
Digital Replicas refer to videos, images, or audio recordings that closely resemble an individual but have been altered from the original. These replicas can be produced using various digital technologies, including AI, and can be either authorized or unauthorized. The First Report highlights that while there are beneficial applications for Digital Replicas – such as accessibility tools and creative projects - there are also significant risks. These risks include job losses for artists, the creation of deepfake pornography, fraud, and the spread of false information.
The First Report points out that while some states have introduced legislation to regulate Digital Replicas, existing federal laws – including the Copyright Act, the FTC Act, the Lanham Act, and the Communications Act - provide only limited protections. Therefore, it emphasizes the need for new federal legislation to adequately safeguard individuals against unauthorized digital copying. The main content of such legislation includes:
(1) Subject of regulation: Highly realistic digital copies that are difficult to distinguish from real-life counterparts are the primary subject of regulation.
(2) Scope of protected individuals: Protected individuals shall include all individuals, not just celebrities or public figures.
(3) Duration of protection: Protection shall be provided during the individual’s lifetime, with limited protection provided after death.
(4) Infringement of rights: Legal liability shall be imposed on unauthorized distribution or disclosure of digital copies, and personal use, including the mere act of creation, shall be excluded from infringement.
(5) Indemnification for online service providers (OSPs): a safe harbor mechanism shall be established for OSPs.
(6) Licensing and Transfer of Rights: Individuals may license their digital reproduction rights, but cannot fully transfer ownership.
(7) Transferability: The right to Digital Replicas involves both privacy interests and property rights. Similar to the right to privacy, full transfer of this right should not be permitted. , Licensing should include safeguards such as time limits or protection of minors.
(8) Consideration of freedom of expression (the First Amendment): A legal framework should be established to ensure that federal laws are balanced with freedom of expression.
(9) Remedy: Strong remedies, including injunctive relief and monetary damages, shall be provided, and provisions for statutorily defined damages and legal fees shall be included.
III. MAIN CONTENT OF THE SECOND REPORT (PART 2: COPYRIGHTABILITY)
The Second Report determined that the existing copyright law is sufficient to address the challenges posed by AI-generated content, indicating that amendments are unnecessary. It is a foundational principle in the U.S. Constitution and judicial precedents is that copyright protection requires human authorship. Consequently, works created solely by AI may not be eligible for copyright protection. However, works that incorporate AI-generated elements may still qualify for copyright if there is a significant creative contribution from human authors.
The Second Report presents the following instances where human intervention may be protected by copyright:
(1) Assistive uses of AI: When AI only assists human creativity in the creative process, rather than replacing it, copyright protection is not affected.
(2) Use of prompts: Simple prompts (commands) do not meet the requirements for copyright protection. Detailed prompts may demonstrate creativity, but they do not constitute direct control over AI’s processing, making it difficult to recognize the user as the author.
(3) Expressive inputs: AI-generated output may be copyrighted if it clearly reflects elements created by humans. For example, if a hand-drawn illustration is input into AI, and such an element is clearly visible in the final output, the original illustration may retain its copyright protection.
(4) Modifying or Arranging AI-Generated Content: If AI-generated content is creatively selected, coordinated, arranged, or modified to achieve originality, human copyright may be recognized.
Regarding the question of extending copyright protection to AI-generated works, the Second Report takes a negative stance. It argues that granting copyright to AI-generated works would undermine human creativity and contradict the fundamental purpose of copyright law. Additionally, the need for extra protection through separate "sui generis rights" for AI-generated content has not been adequately demonstrated.
IV. TOPIC OF THE THIRD REPORT AND RECENT DEVELOPMENTS IN RELATED DISPUTES IN THE U.S.
The Third Report will focus on issues surrounding AI and the utilization of existing works for training, with at least 39 ongoing AI copyright cases currently in the United States alone.1
Recently, a U.S. federal court upheld the plaintiff’s claims in a copyright infringement lawsuit filed by Thomson Reuters against Ross Intelligence.2 The court recognized the plaintiff’s argument that Ross Intelligence had used content from Westlaw, Reuters’ legal database, without permission to train its legal tech AI research platform. While Ross Intelligence contended that its use of Westlaw’s legal case summaries (headnotes) to train its AI constituted a “transformative use” under the fair use principle, the court rejected this assertion. It ruled that Ross Intelligence’s AI merely reorganized Westlaw’s legal research services without imparting any new meaning or interpretation.
Nonetheless, the outcome of this case does not universally apply to all generative AI-related cases. The judgment was based on the key facts that i) Ross Intelligence’s AI uses a method of replicating existing content, which differs from generative AI that learns from diverse data sources to produce entirely new content, and ii) Ross Intelligence’s AI aimed to launch a product that directly competes with Westlaw.
While this case may establish a significant precedent for copyright law concerning how AI learns from and utilizes existing works, it should be viewed separately from discussions regarding the fair use of generative AI.
V. DEVELOPMENTS IN DISPUTES OVER AI AND COPYRIGHT IN KOREA
As the landscape of ongoing lawsuits regarding the use of third-party works for AI training continues to grow in the U.S., Korea recently emerged as the first Asian country to file a lawsuit in this arena.
On January 13, 2025, the three major broadcasting companies (SBS, MBC, and KBS) initiated legal action against Naver in the Seoul Central District Court, seeking to halt copyright infringement and claiming damages for violations of the Unfair Competition Prevention Act. They argued that NAVER's unauthorized use of media companies' data for generative AI training infringes on their copyrights.
As generative AI technology evolves rapidly in Asian markets, it is likely that further legal debate will arise regarding the legitimacy of using third-party works for AI training and the applicability of the fair use principle. Similar disputes are expected to continue in Korea, and the court’s decisions will play a crucial role in shaping the future of AI copyright regulation.
VI. IMPLICATIONS
The series of reports on AI published by the USCO is a key component of its ‘AI Initiative,’ which commenced in early 2023. As part of this initiative, four public hearings and two webinars were conducted, alongside a comprehensive public consultation process that gathered over 10,000 opinions through Notices of Inquiry (NOIs) regarding AI and copyright on August 30, 2023. The insights gained from this extensive research and public feedback are anticipated to serve as vital policy guidelines for delineating the relationship between AI technology and copyright.
Consequently, companies that develop and operate AI technologies, as well as stakeholders in the content industry, should carefully analyze the content and reasoning of these reports in preparation for potential disputes over AI copyright.
In particular, the use of works in the AI training process and the application of the fair use principle are likely to emerge as pivotal issues in forthcoming global and domestic disputes. BKL’s AI Team will continue to monitor ongoing litigation in the U.S., analyzing the rulings to assess their implications for the AI regulatory landscape in Korea, and will provide relevant updates.
[Korean version]
1 The issues and progress of the major litigation filed in relation to OpenAI were covered in the newsletter dated December 17, 2024 entitled "Legal Issues in the Age of AI Revealed through OpenAI-Related Lawsuits".
2 It was rendered as a summary judgment in the U.S. style without going to the main hearing.
