NEW YORK (Diya TV) — A federal court has opened the door for authors to challenge OpenAI over alleged copyright infringement after the judge ruled that authors can pursue claims that go beyond mere training of large language models. The decision marks a significant step in the ongoing battle between writers and the AI industry over how copyrighted works are used in the development of systems like ChatGPT.
In the case brought by a group of authors, including George R.R. Martin, the plaintiffs allege three different types of infringement. First, they say OpenAI trained its AI by using copyrighted books without permission. Second, they claim the company downloaded books from so-called shadow libraries, and without a link to training, that act alone constitutes infringement. Third, they argue that the output generated by ChatGPT is “substantially similar” to the books used in training, making it infringing.
The judge agreed that the shadow library theory can proceed independently from the training theory. He wrote that the fact that many earlier complaints tied illegal downloading to model training “is not dispositive.” In other words, authors may pursue claims that the mere act of copying or storing the works without authorisation can be infringing, even if the works were not used for training the AI.
The court also held that some of ChatGPT’s outputs could be found infringing. For example, in summarising George R.R. Martin’s A Song of Ice and Fire, the chatbot reproduced the tone, plot, characters, and themes in a way the court said a reasonable reader might find “substantially similar” to the original work. The judge emphasised that he is not ruling that OpenAI’s use is fair, but that the authors’ claims are strong enough to go forward.
With these rulings, the case now offers authors multiple routes to victory. If the authors win under any one theory, they may claim statutory damages — up to $150,000 per work for willful infringement. OpenAI may now face not only the training-data claim but also independent claims tied to downloading and to chatbot outputs. For authors, the ruling signals that AI companies may have to pay attention to licensing, data practices, and model output. For OpenAI and the broader AI industry, the decision raises the cost and risk of operating large language models without a clear legal framework for training on copyrighted text.
This decision comes amid a wave of lawsuits by authors, publishers, and news organisations against AI developers. While some courts have sided with tech firms under the doctrine of fair use, others have permitted claims of copying or storing copyrighted works to proceed. The ruling against OpenAI aligns with a growing consensus that the legal boundaries for AI training and output are unsettled — and that rights-holders may prevail in some cases.
