The US government has entered The New York Times’ copyright lawsuit against OpenAI with a 20-page brief supporting the company’s position on using copyrighted material to train large language models without permission.
The filing argues that the United States has an interest in maintaining a competitive AI industry and retaining global leadership in the technology. It warns that applying fair-use doctrine too narrowly to model development could obstruct scientific and creative progress as well as economic opportunity.
That intervention gives OpenAI a powerful institutional ally in the debate, but it does not decide the case. The lawsuit remains before the US District Court for the Southern District of New York, and the brief’s authors have no authority to issue the judgment.
Fair use sits at the center of the dispute
Large language models are trained on enormous collections of published work, including books, articles and other copyrighted media. AI developers have often assembled those collections without obtaining permission from every rights holder. Publishers including The New York Times argue that training on their work without authorization infringes copyright.
The legal answer depends heavily on fair use, which can permit certain unlicensed uses of protected material. In this dispute, a central question is whether training transforms the original works sufficiently to qualify for that exception rather than reproducing or substituting for them.
The government’s brief supports the transformative-use argument. It presents model training as a process that enables different outputs and capabilities, while framing legal restrictions as a potential obstacle to national AI development. That is an advocacy position offered to the court, not a verified description of every training dataset or every model output.
The New York Times’ opposing claim also remains unresolved in this case. TechCrunch’s report describes the legal fault line but does not detail all of the publisher’s allegations, OpenAI’s complete response or the evidence the judge will ultimately weigh.
Earlier litigation offers a qualified signal
Cases concerning AI training and copyright have so far often produced outcomes favorable to AI companies, according to TechCrunch. One prominent decision involving Anthropic illustrates why the distinction between training and data acquisition matters.
Judge William Alsup ordered Anthropic to pay a $1.5 billion settlement to writers whose books were used in training. The penalty, however, concerned the company’s use of illegal shadow libraries to obtain the books, rather than a finding that model training itself was unlawful.
Alsup compared training to an aspiring writer reading existing work in order to create something different. That reasoning supports the idea that learning statistical patterns from copyrighted material can be transformative. It does not establish a universal rule for all AI systems, sources or uses, and a decision in another case does not predetermine the New York litigation.
The distinction creates two separate questions for developers: whether training is legally transformative and whether the material used for that training was acquired lawfully. A company could receive a favorable answer on the first while remaining exposed on the second.
The brief adds influence, not legal certainty
The administration connects its position to an executive order signed by President Donald Trump and to the goal of American leadership in AI. That policy emphasis may affect how forcefully the government presents the economic consequences of a restrictive decision.
Courts, however, interpret copyright law case by case. The district judge is not bound to adopt the administration’s view, and the filing does not eliminate the possibility of later appeals or different conclusions in other disputes.
Its immediate significance is therefore procedural and political. The government has chosen to tell the court that broad room for AI training serves a national interest, strengthening OpenAI’s policy argument alongside its legal defense.
For publishers, model developers and creators, the underlying uncertainty persists. The eventual ruling must address where transformation ends, what rights remain with the owners of training material and whether the way a dataset was assembled changes the result. Until then, the brief shifts the weight of the argument without turning that argument into law.
