George R.R. Martin vs OpenAI: a new era of copyright disputes
When the author of the series “A Song of Ice and Fire”, known to the public through the HBO adaptation “Game of Thrones”, turns his pen into a procedural weapon against OpenAI, legal reality itself seems to have stepped out of fiction.
The case of Martin vs. OpenAI goes beyond the scope of a mere dispute between an established author and a developer of advanced technologies. It stands out as a pivotal moment of legal clarification, as the courts are called upon to determine whether the full reproduction of works during the training process of AI models constitutes a use requiring the authorisation of rights holders or, conversely, whether such reproductions are merely technical in nature, being automatically generated copies devoid of independent economic value.
As held by the United States District Court for the Southern District of New York in its opinia ruling of 27 October 2025, the court rejected OpenAI's request for early dismissal of the case, finding that the claimants' allegations have a plausible legal basis and require a full examination on the merits. This preliminary ruling confirms that disputes concerning the use of protected works as “training material” for algorithmic systems will play a decisive role in shaping the legal framework applicable to artificial intelligence.
I. Procedural background
George R.R. Martin, together with other established authors,[1], brought a class action before the federal court in New York, claiming that their works had been used without authorisation in the process of training the artificial intelligence models developed by OpenAI. At the end of October 2025, the court rejected OpenAI's request for early dismissal of the proceedings, finding that the claimants' allegations were sufficiently well-founded to warrant examination on the merits.
The ruling gives the claimants an important procedural boost and highlights the need to carefully examine how protected works are used in training artificial intelligence.
II. The allegations made
1. Unauthorised reproduction
The claimants argue that OpenAI infringed art. 106 (1) din U.S. Copyright Act (section 106(1) of the U.S. Copyright Act), as their works were allegedly copied in full during the technical processes used to train the models. This allegation concerns the digital reproduction carried out for the purposes of algorithmic processing, even where such copies are not communicated to the public.
2. Creation of derivative works
In the procedural ruling allowing the case to proceed, Judge Sidney Stein held that certain responses generated by ChatGPT may reproduce elements of the plot, characters or themes of the A Song of Ice and Fire series in a manner that could be considered “substantially similar” to the original work. The court emphasised that this is a matter that may be assessed by a jury, without determining at this stage that derivative works had actually been created without permission.
3. Commercial use
A central element of the action is the essentially commercial nature of the technology developed by OpenAI. The claimants emphasise that the language models are integrated into a product exploited commercially, generating direct and indirect revenue, and that this has a decisive bearing on the applicable legal framework.
In its preliminary analysis, the court noted that the commercial nature of the use is one of the essential factors of the fair use, doctrine, set out in art. 107 din (section 107 of the) U.S. Copyright Act. This factor is not, in itself, determinative, but carries greater weight where the use may compete with, substitute for, or affect the current or potential market for the original work. In this context, the Supreme Court, in Andy Warhol Foundation v. Goldsmith (2023), clarified that a commercial use, even one bearing transformative elements, cannot be characterised as fair use where, by its nature and concrete purpose, it conflicts with the licensing market for the original work.
III. OpenAI's defence: fair use in the algorithmic age
In response to these allegations, OpenAI bases its defence on the fair use, doctrine, arguing that the process of training artificial intelligence models is transformativ.
According to the company, the models do not retain or reproduce the works in their original form, but instead operate with abstract statistical representations, used to generate new responses distinct from the source texts. In OpenAI's view, the purpose of this processing is not to compete with the authors' works, but to facilitate technological innovation and the development of advanced language-processing tools.
In a recent ruling, the federal court in California, in the Anthropic PBC case, held that training an artificial intelligence model using protected books may constitute “fair use”, given that the use was found to be transformative and did not directly substitute for the author's work, provided that the data had been lawfully obtained.
In essence, the court held that OpenAI's arguments concerning the transformative nature of the training process and the technical character of the copying could not be assessed without a full examination on the merits, stressing that only the taking of evidence could clarify the real impact on the market and on the authors' rights.
IV. Perspectives from European law:
In the European context, the use of protected works for training artificial intelligence systems is regulated by Directiva (UE) 2019/790 (Directive (EU) 2019/790 - the DSM Directive), transposed into Romanian law by Legea nr. 69/2022 (Law no. 69/2022) amending and supplementing Legea nr. 8/1996 (Law no. 8/1996) on copyright and related rights. The Directive establishes a distinct framework for text and data mining (TDM):
• Art. 3 permits TDM without the rightsholders' authorisation, for the purposes of scientific research, by research organisations and cultural heritage institutions, subject to data security requirements and the retention of copies.
• Art. 4 permits TDM for any purpose, including commercial purposes, provided the rightsholder has not expressly opted out of such use.
Consequently, within the EU, an author who has not consented to such use, or who has expressly opposed it, is entitled to protection, and the commercial use of their work in AI training may give rise to legal issues.
For AI operators providing services within the European area, compliance with these provisions is mandatory. Although Romania has transposed the DSM Directive, the practical application of the TDM regime to complex AI training processes remains partly unsettled, and is likely to be shaped progressively through case law and possible legislative intervention.
V. Legal and economic consequences
Scenario I - An outcome favourable to authors
Should the court find that training AI models on protected works constitutes an infringement of economic rights, the possible consequences would include:
• a requirement to obtain licences for the use of works in training processes;
• the emergence of a collective licensing mechanism, inspired by the model used in the music industry;
• higher operational costs for AI developers;
• the introduction of a specific remuneration system, often referred to as “algorithmic royalties”.
Scenario II - An outcome favourable to OpenAI
Should the court uphold the defence based on fair use:
• the application of this doctrine would be reinforced in the context of emerging technologies;
• AI operators could continue to use materials for training without additional licences;
• authors' bargaining power vis-à-vis technology platforms would diminish;
• the pace of innovation would remain accelerated, albeit with reduced protection for original creators.
Scenario III - A negotiated outcome
An intermediate outcome, and often regarded as the most likely, could consist of adopting a conventional framework reconciling the competing interests:
• collective licensing managed by specialised bodies;
• payment of lump sums to authors;
• mechanisms enabling authors to express their consent or, conversely, their express opposition to the use of their works in training;
• greater transparency regarding the data sources used by AI models.
VI. Conclusion
The Martin vs. OpenAI litigation does not, for the time being, deliver a verdict, but rather marks the opening of a legal front in which copyright and artificial intelligence each lay claim to supremacy.
As courts are called upon to draw the line between reproduction and innovation, between the protection of the work and the power of the algorithm, one may wonder whether, in this digital age, the law will continue to shape the algorithm, or whether the algorithm will ultimately come to reshape the law itself.
[1] John Grisham, Michael Connelly, Arthur Golden, Jodi Picoult, Jonathan Franzen, David Baldacci
Lawyer Ana Vilciu, Buju, Stanciu & Asociații Lawyer Maria Vaduva, Buju, Stanciu & Asociații