GenAI's Copyright Conundrum

ChatGPT’s rendition of Studio Ghibli-style art, generated within 2 minutes. Image credits: the art style of Hayao Miyazaki (creator of Studio Ghibli) generated by ChatGPT

Imagine commissioning Studio Ghibli to create a custom animation for you, free of charge. But instead of Hayao Miyazaki, the amazing creator behind the iconic Japanese animations, the artist is Artificial Intelligence. 

As we admire AI-generated masterpieces making their rounds online, something does not sit right: Is this even ethical? Does this not diminish the value of human artist’s dedication and originality in their craft?

Understanding Copyright Laws

"The legal right to control all use of an original work, such as a book, play, movie, or piece of music, for a particular period of time" ~Cambridge Dictionary

The history of copyright laws can be traced to the British Statute of Anne 1710, which sought to prevent the unauthorised republication of printed books by third parties. This coincided with the rise of the printing press – any form of copying prior was by hand, rendering copying virtually impossible – which allowed for mass manufacturing and reproduction of original written creations. 

Since then, copyright laws have periodically evolved based on prevailing technological inventions and legal disputes involving reproduction of original creations. 

Copyright protection aims to reward authors through granting exclusive property rights. These protections incentivise innovation by rewarding effort in name and through economic benefits, ultimately enriching society through new ideas in science and the arts.

At this point, It is crucial to distinguish that ideas and facts are not protected by copyright. Instead, it is the expression of these that are eligible for protection.

Copyright and AI-generated creations inspired by original creations

Outright copying, such as lifting a melody or plagiarising written content is clearly illegal under most copyright laws. 

When it comes to “inspiration”, however, we stand in the grey area of copyright laws. Generative AI relies entirely on databases built from third-party materials. While some companies seek formal usage rights to copyrighted content in training, AI itself cannot create without input.

Yet, it would be inaccurate to label all AI-generated content as infringing. If you think about it,  human thoughts and opinions are formulated based on our education received from copyrighted sources like books and articles. Yet, these are not considered copyright violations, as they often involve facts or broadly accessible information.

In that same breadth, how would this “inspiration” now apply to AI?

This tension lies at the heart of current legal debates. In an ongoing case between major music publishers, including UMG and Concord Music Group, and Anthropic PBC, the music publishers have alleged that Anthropic’s AI chatbot, Claude, infringes copyright by generating lyrics based on a dataset including their copyrighted songs. Yet, Anthropic countered that Claude does not reproduce original lyrics. No final judgements have been made till date, but the US court has denied the plaintiffs’ request for a preliminary injunction, citing no immediate irreparable harm. This allows Claude to continue its function as the case unfolds.

This case — along with AI-generated content such as the case of Studio Ghibli — raises a broader issue: Can creative styles be copyrighted?

Legally, the answer is no. As with musical genres, artistic styles cannot be protected by copyright. This principle was reaffirmed by the U.S. Supreme Court in Google v. Oracle, which stated:

“Copyright protection cannot be extended to any idea, procedure, process, system, method of operation, concept, principle, or discovery... Unlike patents, which protect novel and useful ideas, copyrights protect expression, but not the ideas behind it.”

As committed editors, we had to try the generation out ourselves first: it only took a minute to generate exactly what we had in mind. Convenient? Yes. Ethical? Not really... 

Still, ethical concerns remain. The ability to replicate the creations of established creators in minutes using AI continues to provoke debate about fairness in the age of AI-generated creativity. Next, we will take a look at issues that have arose from the use of copyrighted content in training of Large Language Models.

TDM: Navigating Legal Boundaries 

As most AI systems (or more specifically large language models that interpret human data) are trained on huge data sets i.e. the text and data mining (TDM) process, conflict over the use of copyrighted content has emerged as a major point of contention. 

In this section, we'll explore how the US and EU have adopted varying approaches to this issue. 

Following the creation of the British Statute of Anne 1710, the U.S. introduced the Fair Use Doctrine alongside copyright laws. Lawmakers recognised that overly strict protections could stifle further innovation and expression needed for societal progress. Fair use addresses this as an affirmative defense point raised in response to a copyright claim, permitting limited use of copyrighted material without permission under certain conditions, such as news reporting, commentary and criticism, and education (as we are doing in this article).

While fair use itself lacks cemented rules unlike laws, 4 main factors are usually used to evaluate fair use in the US.

To understand how fair use is applied in the US, we can look at the New York Times v. OpenAI case. NYT accused OpenAI of utilising their articles to train AI models like ChatGPT, raising copyright concerns. Although the case is ongoing, US courts have decided to proceed with hearings given the current importance to provide further clarifications on this copyright concerns.

In this January’s hearing, NYT lawyers presented evidence reflecting ChatGPT’s reproduction of NYT articles verbatim when asked about related topics. The court’s key consideration was on the transformative-use factor under fair use, arguing that ChatGPT failed to make meaningful changes to the original content. 

Economically, the market-substitution factor also weighed in, as ChatGPT’s responses could harm NYT's market by replicating its exclusive content.

While we await clearer legal regulation in the US, media companies have begun acknowledging the challenges in preventing AI developers from accessing publicly available content. Instead of resisting, some—like Dow Jones (parent of The Wall Street Journal)—have launched content licensing marketplaces. These platforms allow corporations to license content for AI use, promoting revenue-sharing and mutual recognition of both original creators and AI developers. 

Across the Atlantic, the EU has also published legally-binding rules outlining the use cases of copyrighted material for LLM training. These rules are scheduled to be fully implemented across all member states by 2021. 

Directive 2019/790 (also known as the Copyright Directive) introduces text and data mining (TDM) exceptions that allow limited uses of copyrighted material for LLM training purposes. 

More specifically: 

• Article 3 allows TDM for “scientific and research purposes” with no opt-out is permitted. Notably, this exception is only permitted when developers of AI systems have “lawful access” i.e. through subscriptions or open-access licenses.  

• Article 4 allows TDM for any other purpose that includes commercial uses. However, copyright holders can request to opt out of TDM and the request must be respected by developers of AI systems. 

Entering into 2024, provisions of EU's AI Act have offered further clarifications on the bloc's stance. For instance, Art. 53(1)(c) reiterates that developers must respect opt-out requests as per the Copyright Directive's article 4. In addition, Art. 53(1)(d) creates a transparency obligation that all developers of AI systems must abide by - they must reveal the sources of data used to train AI systems, thereby enabling rightholders to easily opt-out should they wish to do so. 

In summary, the treatment of TDM practices best exemplifies the differing approaches to copyright law across major jurisdictions. In the U.S., this is handled flexibly through the fair use doctrine that considers factors like purpose and market impact on a case-by-case basis. In contrast, the EU has adopted a stricter, rules-based approach that leaves less room for ambiguity.

Now that we’ve looked at the legal issues arising from the use of copyrighted material to train AI systems, let’s next explore whether the content created by AI can be granted by copyright protection. 

At present, human authorship is a fundamental requirement for a work to be eligible for protection under the copyright law of multiple jurisdictions. We will now briefly cover how major jurisdictions interpret the issue of copyright protection for AI-generated works. 

Under US law, this macaque does not own copyright to a selfie that it took by accident. Image credit: CNN

Back in 2014, a judge ruled in Naruto v Slater that a monkey cannot claim copyright ownership. This principle was reaffirmed in the case involving Dr Thaler’s AI system DABUS, whose artwork A Recent Entrance to Paradise was denied copyright protection due to the absence of human involvement. 

"absent any human involvement”, which is itself a “bedrock requirement of copyright" ~Judge Beryl A. Howell, US District Court for DC commenting on the DABUS case

Similarly in Zarya of the Dawn, author Kris Kashtanova was granted copyright protection over the plot and image arrangement of the work, but not the individual images that were generated entirely using AI system Midjourney.

Since the conclusion of Zarya of the Dawn, the U.S. Copyright Office has further clarified its stance. Applicants must now disclose the presence of any AI-generated material and explain the human contribution involved prior to registering their works as part of the copyright registration process. 

Notably, the US has maintained that merely providing prompts to an AI system is not enough to claim authorship, given that prompts (no matter how detailed) function more as instructions or ideas, which are not protectable under copyright law. 

That said, derivative authorship is recognised under US law. If a human modifies or arranges AI-generated material in a way that introduces new, original expression, the resulting work may qualify for copyright. In such cases, copyright protection will only extend to the human-created elements, not to the underlying AI-generated content (as seen in the Zarya of the Dawn case).

Moreover, the inclusion of AI-generated components within a larger, human-authored work (e.g a film that uses AI-generated special effects) will not disqualify the overall work from copyright protection.

The European Union adopts a similar approach to copyright protection. Works entirely determined by technical processes or lacking meaningful human intervention are unlikely to qualify. The European Court of Justice defines a copyrightable work as the “author’s own intellectual creation” that contains the author’s “personal touch”. In rejecting Dr Thales’ patent application for DABUS, the European Patent Office reasoned that inventors must be natural persons. 

In contrast, China has taken a markedly different approach regarding copyright protection for AI-generated works. As illustrated in Shenzhen Tencent v. Shanghai Yingxun, copyright protection was granted to Tencent for a financial report generated by its Dreamwriter software. 

Notably, the court held that the report demonstrated intellectual effort as it reflected the “team’s personalised choice, judgment, and skills” and “exhibited a certain degree of originality” on the part of the Tencent team, thereby warranting protection under Chinese copyright law. In essence, Chinese courts have established that the presence and degree of human contribution are predominant factors in determining the copyrightability of AI-generated works.

However, a notable exception is the United Kingdom's Copyright, Designs and Patents Act 1988 (CDPA) that explicitly provides copyright protection for computer-generated LDMA works.

Section 9(3), CDPA: The author of a computer-generated LDMA (literary, dramatic, musical or artistic) work, “shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken”. 

In theory, the CDPA can be invoked in the future to attribute authorship of AI-generated works to AI developers. Even then, uncertainties remain: the UK’s Court of Appeal has recently interpreted the CDPA’s “originality” requirement in THJ Systems Ltd v Sheridan to follow the higher standard as set by the EU test of “author’s own intellectual creation” (that introduces a creative element).

A Sui Generis Approach to Copyright for AI-generated works

As AI systems become increasingly complex, many believe that offering some form of copyright protection for AI-generated works will help to promote investment in AI systems. For instance, legal clarity that would allow for commercial licensing of content output will provide developers with a financial incentive for continued research and innovation of their AI systems.

In light of the dynamic nature of GenAI, some academics have proposed creating a sui generis (Latin for “of its own kind or class”) right for AI creations (rather than relying on the cumbersome process of updating existing copyright law).

A sui generis system would grant protection to works that presently fall outside of conventional IP law. Only AI works with human contributions will fall under the purview of this new framework. Works generated autonomously by AI systems should remain in the public domain and can be freely used by anyone. 

Under the sui generis framework, only economic rights will be granted to right-holders for a limited period of time. The nature of works created with the involvement of GenAI also means that sui generis right-holders will only be afforded protection from works of verbatim similarity. 

The evolution of sophisticated AI systems has led to the question of how copyright protection should apply to AI-generated works (if at all) emerging at the forefront of the debate across legal circles. Traditional copyright law based on the principles of human authorship and originality have struggled to apply to AI creations that are often derived from existing works or data. The flexibility afforded by a sui generis framework granting AI-generated works limited economic rights has emerged in the meantime, allowing for a practical balance to be struck between innovation and copyright protection while safeguarding human creativity. 

References

Abdel, A., Ismail, M., Massadeh, F., & Alnusair, F. (2024). The legal protection of artificial intelligence-generated work: The argument for sui generis over copyright. Corporate Law & Governance Review, 6(1), 49–56. https://doi.org/10.22495/clgrv6i1p5

Allyn, B. (2025, March 26). Judge allows “New York Times” copyright case against OpenAI to go forward. NPR. https://www.npr.org/2025/03/26/nx-s1-5288157/new-york-times-openai-copyright-case-goes-forward

Bond, T., & Blair, S. (2019). Artificial Intelligence & copyright: Section 9(3) or authorship without an author. Journal of Intellectual Property Law & Practice, 14(6), 423. https://doi.org/10.1093/jiplp/jpz056

Bottomley, D. (2024). A Comparative Analysis of the need for sui generis Artificial Intelligence Legislation in Kenya and South Africa. https://cipit.strathmore.edu/wp-content/uploads/2024/09/A-Comparative-Analysis-of-the-need-for-sui-generis-Artificial-Intelligence.pdf

Brittain, B. (2025, March 26). Anthropic wins early round in music publishers’ AI copyright case. Reuters. https://www.reuters.com/legal/anthropic-wins-early-round-music-publishers-ai-copyright-case-2025-03-26/

Cambridge Dictionary. (2020, January 29). COPYRIGHT | meaning in the Cambridge English Dictionary. Cambridge.org. https://dictionary.cambridge.org/dictionary/english/copyright

Chew, J., & Davidson, J. (2025, November 1). The interaction between intellectual property laws and AI: Opportunities and challenges. Nortonrosefulbright.com. https://www.nortonrosefulbright.com/en-us/knowledge/publications/c6d47e6f/the-interaction-between-intellectual-property-laws-and-ai-opportunities-and-challenges

Copyright Alliance. (2016, September 7). What Is Fair Use? Copyright Alliance. https://copyrightalliance.org/faqs/what-is-fair-use/

Copyright Alliance. (2020, November 4). What Is The Purpose of Copyright Law. Copyright Alliance. https://copyrightalliance.org/education/copyright-law-explained/copyright-basics/purpose-of-copyright/

Copyright timeline: A history of copyright in the united states - association of research libraries. (2019). Association of Research Libraries; ARL. https://www.arl.org/copyright-timeline/

Cordell, N., Potts, B., & Dickens, R. (2024, August 20). Ownership of AI-generated content in the UK. Aoshearman.com; A&O Shearman. https://www.aoshearman.com/en/insights/ownership-of-ai-generated-content-in-the-uk

Dykema. (2025). The Future of Creativity: U.S. Copyright Office Clarifies Copyrightability of AI-Generated Works. Dykema - the Future of Creativity: U.S. Copyright Office Clarifies Copyrightability of AI-Generated Works. https://www.dykema.com/news-insights/the-future-of-creativity-us-copyright-office-clarifies-copyrightability-of-ai-generated-works.html

Guadamuz, A. (2018, February 21). Can the monkey selfie case teach us anything about copyright law? WIPO Magazine. https://www.wipo.int/web/wipo-magazine/articles/can-the-monkey-selfie-case-teach-us-anything-about-copyright-law-40287

Hardman, B., & Housel, J. (2023, August 30). A Sui Generis Approach to the Protection of AI-Generated Works: Balancing Innovation and Authorship. Social Science Research Network. https://doi.org/10.2139/ssrn.4557004

Hugenholtz, P. B. (2019, July 24). The New Copyright Directive: Text and Data Mining (Articles 3 and 4). Kluwer Copyright Blog. https://copyrightblog.kluweriplaw.com/2019/07/24/the-new-copyright-directive-text-and-data-mining-articles-3-and-4/

Intellectual Property Owners Association. (2020). Sui Generis Right for Trained AI Models. https://ipo.org/wp-content/uploads/2020/11/SG-model-rights-committee-paper-pub.pdf

Kucukali, B. (2022). The protection of AI-generated works under european copyright law: toward adoption of a neighbouring rights approach. The University of British Columbia. https://open.library.ubc.ca/media/stream/pdf/24/1.0421298/4

Manteghi, M. (2024, May 18). Will the AI Act Bring More Clarity to the Regulation of Text and Data Mining in the EU? EU Law Analysis. https://eulawanalysis.blogspot.com/2024/05/will-ai-act-bring-more-clarity-to.html?m=1

Naruto v. Slater, No. 16-15469 (9th Cir. 2018). (n.d.). Justia Law. https://law.justia.com/cases/federal/appellate-courts/ca9/16-15469/16-15469-2018-04-23.html

Novelli, C., Casolari, F., Hacker, P., Spedicato, G., & Floridi, L. (2024). Generative AI in EU Law: Liability, Privacy, Intellectual Property, and Cybersecurity. Social Science Research Network. https://doi.org/10.2139/ssrn.4694565

Rivaux, S., Fink, M. F., & Leonard, M. (2024). Anthropic Tries to Strike a Chord with the Music Industry. Pillsbury Law. https://www.pillsburylaw.com/en/news-and-insights/anthropic-copyright-claude-ai.html

Sun, H. (2022). Redesigning Copyright Protection in the Era of Artificial Intelligence. Iowa Law Review. https://ilr.law.uiowa.edu/sites/ilr.law.uiowa.edu/files/2023-02/A6_Sun.pdf

UK Intellectual Property Office. (2022, June 28). Artificial Intelligence and Intellectual Property: copyright and patents: Government response to consultation. GOV.UK. https://www.gov.uk/government/consultations/artificial-intelligence-and-ip-copyright-and-patents/outcome/artificial-intelligence-and-intellectual-property-copyright-and-patents-government-response-to-consultation#copyright-in-computer-generated-works

Voytsitska, V. (2024, June 28). How can the opt-out mechanism under article 4(3) of directive (eu) 2019/790 (cdsmd) be optimised to balance the copyright protection of the news articles with the advancement of the text and data mining (tdm) across the european union. Universiteit Utrecht. https://studenttheses.uu.nl/bitstream/handle/20.500.12932/48186/Thesis%20Viktoriia%20Voytsitska.pdf

Weatherbed, J. (2025, January 3). Anthropic reaches deal with music publishers over lyric dispute. The Verge. https://www.theverge.com/2025/1/3/24334866/anthropic-claude-music-publishers-lyric-copyright-lawsuit-deal

White & Case. (2024, March 31). AI Watch: Global regulatory tracker - United States | White & Case LLP. White & Case. https://www.whitecase.com/insight-our-thinking/ai-watch-global-regulatory-tracker-united-states

Wolfson, S. (2023, March 23). The Complex World of Style, Copyright, and Generative AI. Creative Commons. https://creativecommons.org/2023/03/23/the-complex-world-of-style-copyright-and-generative-ai/

Zhou, B. (2019). Artificial Intelligence and Copyright Protection --Judicial Practice in Chinese Courts. World Intellectual Property Organization. https://www.wipo.int/export/sites/www/about-ip/en/artificial_intelligence/conversation_ip_ai/pdf/ms_china_1_en.pdf

Zhuk, A. (2023). Navigating the legal landscape of AI copyright: a comparative analysis of EU, US, and Chinese approaches. AI and Ethics, 4. https://doi.org/10.1007/s43681-023-00299-0