Copyright Holders and Artificial Intelligence: A Never-Ending Battle

By Pryanna Pradhan '28

---

As artificial intelligence (AI) continues to develop, companies' appetite for data seems endless. To satiate the need for training data, companies sacrifice consumer privacy, bypass copyright laws, and hide behind the premise of fair use. As stated in the landmark case Fiest Publications, Inc. v. Rural Telephone Services Co., copyright is meant to permit “others to build freely upon ideas and information conveyed by a work.” (1) The nature of copyright is meant to preserve the exchange of knowledge; therefore, actual copyright law tends to lack specificities when dealing with AI. For instance, according to copyright law, the AI-generated images that obviously imitate the style of artists have no basis for an infringement case since there is no way to strictly define a style. (2) Though artists can attempt to find workarounds using other common laws, the Copyright Office concedes that the law is insufficient in this area, partially due to the rapid development of AI. (3) Created under the assumption of a human author, qualifications for fair use, such as whether or not the infringer had access to the initial work, become complicated to prove with AI. (4)  Due to examples like these, rulings are hurting copyright owners. Understanding where fair use falls short in author protection remains critical to balancing the interests of the copyright holder as well as the development of AI. To ensure ethical innovation, there must be changes to the AI regulations in place now, with an emphasis on specific and technical rules.

Fair Use Analysis

When a plaintiff cites “fair use” as justification for the use of a copyrighted work, courts must use Section 107 of the Copyright Act to conduct a fair use analysis. (5) The court considers:

1. Purpose of use.

2. Nature of copyrighted work.

3. Amount of copyrighted work used.

4. How much a new work impacts the market of the copyrighted work. (6)

In the next case, it becomes apparent that the purpose of use, as well as the amount of work used, can be fairly applied to the training of AI models on copyrighted works. When it comes to the nature of copyrighted work and how much a new work impacts the market, however, it becomes more complicated to prove. 

The case, Bartz v. Anthropic, highlights the intricacies of determining fair use when AI is involved. (7) A group of book authors argued that Anthropic PBC, the developers of Claude, violated copyright law by using copyrighted works as training material. Anthropic utilized copyrighted books as an internal permanent library and to train large language models (LLMs). After conducting the fair use analysis, the court decided that training the LLMs on copyrighted data and digitizing printed copies into the library had a fair purpose, proper account, and sustainability, and therefore, it was ruled fair use. On the other hand, pirating the works and then placing the copies in the library was decided to violate copyright laws. (8) 

The Case Against the Market Dilution Argument

It may be worthwhile to more closely analyze one of the claims the plaintiffs in Bartz v. Anthropic used. They cited the “market dilution” phenomenon, which states that though the outputs created by AI may not directly mimic the copyrighted work, the AI works satisfy the demand of the market and harm the copyright holder's income. (9) 

Unfortunately, market dilution is a fallible argument against the backdrop of copyright law. Even though LLMs are not pushing out direct replicas of the content they are trained on, market dilution believes that the models creating outputs inspired by their work will cause competition in the market. Setting aside the sheer quantity of works AI models can generate compared to a human, this phenomenon seems to be no different than new, upcoming artists using copyrighted paintings as a point of inspiration, as explained in the Bartz v. Anthropic ruling. (10) 

In fact, many previous copyright cases have discussed similar situations across a diverse range of contexts. For instance, in Google LLC v. Oracle, legal issues arose because Google utilized some of Oracle’s programming language and Application Programming Interface in its own creation. (11). The court ruled, under copyright law, that new creators can still explicitly use parts of the code in their own applications, as Google did. (12)  Or in the case of Cariou v. Prince, the plaintiff, Patrick Cariou, accused Richard Prince, the defendant, of copyright infringement, as Prince utilized many of Cariou’s photographs in his exhibitions. (13) The Court of Appeals ruled that because Prince’s work was transformative, meaning, specifically, the "composition, presentation, scale, color palette and media” are completely different from the photographs, the infringement claim was denied. (14) In both cases, though the demand of the copyright holders could be harmed, it is only considered “harmed” if it violates the explicit guidelines of copyright, which did not occur in either case. 

While market dilution is a real phenomenon, using precedential cases, it is not a sufficient argument against fair use. However, it could be argued that mitigating the impact of market dilution lies not in the interpretation of fair use, as rules against it inherently violate the basis of copyright. 

The Digital Millennium Copyright Act

The Digital Millennium Copyright Act (DMCA) provides a strong foundation to explore a solution. Similar to technological innovation now, in 1998, this act amended copyright law to accommodate the internet. It created the notice-and-takedown system, allowing copyright holders to report potential infringement to a provider and request that the content be taken down. It also established infringement consequences to protect copyright holders, as well as online copyright protection. (15) The DMCA completely transformed how users and owners interact with copyright; as the internet increased the ease of utilizing copyrighted works, the U.S. Copyright Office took measures to preserve copyright, rather than altering the law itself.

The DMCA implements methods that still respect the core of copyright, adjusting regulations around the technology itself. Its strength lies in the emphasis on implementing practices that encourage copyright holders to share access to the works. (16) In the age of AI, companies and copyright holders need rules like these to protect the interests of society.

Potential Regulations

Using the DMCA, there can be an attempt to create a system that protects copyright authors against generative AI. By creating these regulations, copyright holders place preventative measures against generative AI blatantly copying style and establish citations for AI models that are trained on their works. Since using copyrighted works for training AI models is already deemed fair use, by implementing regulations that give credit back to the holders, pushback can be minimized.  

A potential regulation stems from Section 1201 of the DMCA, which prohibits consumers from finding workarounds on protected works. (17) Though companies try to ban prompts explicitly asking to create an output that replicates a style, users easily find prompts to bypass these restrictions. For instance, in 2025, consumers discovered ChatGPT could create images in the style of Studio Ghibli, and the feature quickly became mainstream. (18) Soon after, OpenAI released a content restriction, banning users from explicitly asking the chatbot to create an image in the style of Studio Ghibli. (19) Unfortunately, this measure was futile. Even with that measure in place, there were influxes of articles and advice on specific prompts users could feed into ChatGPT to recreate an image in the style of Studio Ghibli, bypassing the restrictions set in place. From this example, it follows that there should be a regulation prohibiting users from explicitly asking for output that directly copies the style of copyrighted work, plus an additional check for style similarity before the image is given to the user. If the similarity is too high, then the content will not be provided to the user. The style similarity metric can mitigate the issues of skirting the rules AI companies have put in place, while also protecting the original author’s interest. This decreases direct market competition for authors and will avoid undermining their efforts.

Alongside the direct ban on asking for a creation in a specific style, and the style similarity metric, understanding the Copyright Alternative in Small-Claims Enforcement (CASE) Act could inspire more viable protections for holders. This act, created by the Copyright Claims Board, takes claims of copyright infringement, noninfringement, notices, and counter-notices and reviews them, checking for validity and following through with the next appropriate steps. (20) Applying to AI outputs, if an author finds a product that seems to violate their copyright, they can report it to this board and take the next plausible steps. In this way, if the ban and similarity metric fails, there is still another protection holders can fall back on.

Shifting to regulations the AI companies should enact, companies can create publicly available documents that indicate the works used to train their model. This way, authors are aware if their work is used to train an AI model. In a similar vein, to ensure public awareness, mandating that every AI-generated output must contain a permanent watermark or stamp indicating AI generation. Overall, these solutions may not completely stop market dilution, but they will ensure consumers are aware of which were AI-generated, and which were human-made. Comprehensively, these rules may reinforce the importance of ownership and originality in the consumer’s mind, making them more mindful as they generate works from AI. 

Conclusion

Owning a creative work has never been so weakly protected. Intellectual property is threatened at every cornerstone of AI development, and to protect creators, as well as technological development, increased regulation is required. At this moment, copyright cannot protect against market dilution or style plagiarism, and authors pay the price. In the long term, continued copyright infringement met with inaction will result in the loss of human-generated creativity. Luckily, the DMCA and the CASE Act prove that if expectations are established, AI companies, copyright holders, and the law can work together to preserve everyone’s best interests. Placing specific style bans, creating a style similarity metric, implementing a notice-and-takedown system, and providing documents with training data are all steps in the right direction to achieve this vision. With the correct implementation, these regulations will ensure copyright holders maintain their rights to protect their creativity, preserving their ability to express themselves. 

Endnotes

1. Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991).

2. United States Copyright Office. Copyright and Artificial Intelligence, Part 1: Digital Replicas. Washington, DC: U.S. Copyright Office, 2025. 

3. Ibid.

4. Christopher T. Zirpoli. Generative Artificial Intelligence and Copyright Law. LSB10922. Washington, DC: Congressional Research Service, September 29, 2023. Congress.gov. 

5. U.S. Copyright Office. “U.S. Copyright Office Fair Use Index”. Copyright.gov. Last modified August 2025. https://www.copyright.gov/fair-use/index.html.

6. Ibid.

7. Bartz v. Anthropic PBC, 787 F. Supp. 3d 1007 (N.D. Cal. 2025). 

8. Ibid.

9. Brandon Bulter. “The Copyright Office on Fair Use and AI.” re:create, May 16, 2025. https://recreatecoalition.org/the-copyright-office-on-fair-use-and-ai/. 

10.  Bartz v. Anthropic PBC, 787 F. Supp. 3d 1007 (N.D. Cal. 2025).

11. Google LLC v. Oracle America, Inc., 593 U.S., 141 S. Ct. 1183 (2021).

12. Ibid.

13.  Cariou v. Prince, 714 F.3d 694 (2d Cir. 2013).

14. Ibid.

15. U.S. Copyright Office. “The Digital Millennium Copyright Act.” Copyright.gov. 

16. Ibid.

17. Ibid.

18. Grant Slatton. “OpenAI Is Getting Overwhelmed by ‘Ghiblified’ Photo Edits. So Is the Guy Who Helped Popularize the Craze.” Business Insider, March 29, 2025. https://www.businessinsider.com/ghibli-ai-trend-viral-response-heartwarming-2025-3

19. Ibid.

20. United States Copyright Office. “Copyright Small Claims and the Copyright Claims Board.” https://www.copyright.gov/about/small-claims/.