OpenAI Loses Bid to Dismiss Multi-District Class Action Over ChatGPT Outputs

A New York judge ruled on Monday that OpenAI cannot stop a consolidated, multi-district class action brought against by dozens of authors for direct copyright infringement by the outputs of its large language model (LLM), ChatGPT.

OpenAI argued that the plaintiffs had failed to allege substantial similarity between the works and ChatGPT’s outputs, but Judge Sidney Stein of the U.S. District Court for the Southern District of New York said  that “[a] more discerning observer could reasonably conclude that the allegedly infringing outputs are substantially similar to plaintiffs’ copyrighted works.”

The “more discerning observer” test applies when the works in question “contain both copyrightable elements—such as characters, plot, and setting…—and non-copyrightable elements—such as stock themes, stock characters, scenes a faire, and abstract ideas,” explained the court. Under that test, “the relevant question is whether there exists ‘substantial similarity between those elements, and only those elements, that provide copyrightability to the allegedly infringed [work].’” The opinion also said in a footnote however that it would have reached the same conclusion applying the less refined “ordinary observer” test.

The plaintiffs included examples incorporated by reference from the Consolidated Class Action Complaint in their opposition to the motion to dismiss of allegedly infringing outputs such as summaries of “A Game of Thrones” author George R.R. Martin’s works which, despite not recounting “‘[e]very intricate plot twist and element of character development’….most certainly attempts at abridgment or condensation of some of the central copyrightable elements of the original works such as setting, plot, and characters.”

Judge Stein replicated the ChatGPT output of several summaries of Martin’s works and concluded that “the summary conveys the overall tone and feel of the original work by parroting the plot, characters, and themes of the original.” Stein also addressed in a footnote OpenAI’s argument that its outputs are “analogous to the summaries of news articles that the Court determined were not substantially similar to copyrighted works in its decision in N.Y. Times Co. v. Microsoft Corp., 777 F. Supp. 3d 283 (S.D.N.Y. 2025),” distinguishing the outputs in N.Y. Times because they “merely summarized non-copyrightable elements of the original news article,” such as facts, and “differed ‘in style, tone, length, and sentence structure’” from the originals.

Stein ultimately found that a reasonable jury could undoubtedly determine that that these outputs are “substantially similar to Martin’s original work based on the output’s incorporation of such copyrightable elements of Martin’s original work as setting, plot, and characters,” and said the Martin examples were sufficient to defeat OpenAI’s motion to dismiss.

Stein concluded by clarifying that the opinion does not represent a view as to whether the outputs are protected under the doctrine of fair use, which courts generally do not address until the summary judgment phase.

Image Source: Deposit PHotos
Image ID: 640750100
Author: Skorzewiak