IP Law Daily, COPYRIGHT—S.D.N.Y.: OpenAI unable to dismiss authors’ infringement claims based on ChatGPT-generated outputs, (Oct 29, 2025)

By Robert Margolis, J.D.

Applying “more discerning observer” test, court determines reasonable jury could find that outputs (summaries and outlines) are substantially similar to authors’ copyrighted works.

A group of authors, suing on behalf of a class of similarly situated authors, has stated a claim for copyright infringement against OpenAI, Inc. and several related defendants (collectively, “OpenAI”), based on outputs generated by OpenAI’s ChatGPT, the federal district court in New York City has held. The court denied OpenAI’s motion to dismiss the authors’ infringement claims alleging that outputs from ChatGPT, in particular, summaries and outlines of the authors’ copyrighted works, infringe the authors’ works. After reviewing ChatGPT-generated summaries and outlines of Game of Thrones author George R.R. Martin’s works, the court held that a reasonable jury could find those summaries to be substantially similar to the original under the “more discerning observer” test (In re OpenAI, Inc. Copyright Infringement Litigation, No. 1:25-md-03143-SHS-OTW (S.D.N.Y. Oct. 27, 2025)).

The case is the result of the consolidation of several putative class actions against OpenAI and Microsoft Corporation. Authors and copyright holders of fiction and non-fiction books allege that OpenAI and Microsoft infringed plaintiffs’ copyrights in their books by downloading and reproducing their works, using the works to train OpenAI’s artificial intelligence (“AI”) large language models (“LLMs”), and then creating outputs from OpenAI’s LLM products that infringe the authors’ works. OpenAI moved to dismiss claims alleging that the ChatGPT outputs are infringing works.

Substantial similarity. The authors’ consolidated complaint references but does not attach ChatGPT-generated summaries and outlines of several authors’ works. They attached representative outputs to their brief in response to the motion to dismiss. The court held that it could consider those attachments to the brief in determining whether the authors could state a claim for infringement based on those outputs, rejecting an argument by OpenAI. It then reviewed the outputs and found under the “more discerning observer” test for substantial similarity, that the authors stated a claim.

A prima facie claim for copyright infringement requires a showing that (1) the defendant actually copied the plaintiff’s work, and (2) a substantial similarity exists between the defendant’s work and the protectable elements of the plaintiff’s work. Here, there was no dispute over the first element, so the court addressed whether the ChatGPT-generated outputs are substantially similar to the authors’ copyrighted works. And because the authors’ copyrighted works contain both protectable and non-protectable elements,the court applied the “more discerning observer” test to determine substantial similarity. Under that test, the court asks whether there is substantial similarity between only the protectable elements of the parties’ respective works, i.e., treatment, details, scenes, events, and characters. Stock themes, stock characters, and “scenes a faire” do not enjoy copyright protection so are not considered in the “more discerning observer” test. When applying the test, courts in the second circuit do not “dissect” the works at issue into their constituent parts, but rather compare the works’ “total concept and overall feel,” using “good eyes and common sense.” Peter F. Gaito Architecture, LLC v. Simone Dev. Corp., 602 F.3d 57, 66 (2d Cir. 2010).

Applying that test, the court found substantial similarity between ChatGPT generated summaries and outlines of the authors’ works, on the one hand, and the works themselves, on the other. It cited as illustrative samples a ChatGPT-generated summary of Martin’s Game of Thrones and a ChatGPT-generated outline of his works. As to both, the court found that a more discerning observer could conclude that the output is substantially similar to the original author’s work. The summary conveyed the overall tone and feel of the original work, as it “parrot[ed]” the original’s plot, characters, and themes. Similarly, the outline incorporated such copyrightable elements of Martin’s original work as its setting, plot, and characters. Finding it sufficient from those two outputs to determine that the authors state a claim based on ChatGPT outputs, the court did not consider other examples proffered by the authors.

Fair use. Though it denied the motion to dismiss, based on finding the authors stated a prima facie case based on ChatGPT outputs, the court made clear that nothing in its opinion is intended to suggest a view on whether the allegedly infringing outputs are protected by the fair use doctrine.

The Case is No. 1:25-md-03143-SHS-OTW.

Judge: Stein, S.

Attorneys: Anna Josefine Freymann (Lieff Cabraser Heimann & Bernstein, LLP) for David Baldacci. Allison S. Blanco (Latham & Watkins LLP) for Open AI Inc., OpenAI Opco LLC and OpenAI GP LLC.

Companies: OpenAI, Inc.; OpenAI OpCo LLC; OpenAI GP, LLC; OpenAI, LLC; OpenAI Global LLC; OAI Corporation LLC; OpenAI Holdings LLC; OpenAI Startup Fund I LP; OpenAI Startup Fund GP I LLC; OpenAI Startup Fund Management LLC; Microsoft Corp.

MainStory: TopStory AINews Copyright TechnologyInternet NewYorkNews GCNNews