In Re OpenAI Inc., Copyright Infringement Litigation

District court denies motion to dismiss plaintiffs’ claim of direct copyright infringement based on AI outputs, holding that plaintiffs adequately pled that defendant OpenAI actually copied plaintiffs’ text and that reasonable jury could find that allegedly infringing outputs are substantially similar to plaintiffs’ works.

Plaintiffs are authors and copyright holders of fiction and nonfiction books who brought copyright infringement claims against OpenAI Inc., certain of its affiliated companies and Microsoft Corp. The action resulted from the consolidation of multiple putative class actions against OpenAI and Microsoft into a multidistrict litigation (MDL) that was centralized before the court earlier this year. (Read several of our summaries of earlier decisions in the original actions here, here, here and here and in the multidistrict litigation, here.) Plaintiffs allege that defendants infringed plaintiffs’ copyrights by downloading and reproducing plaintiffs’ works, using those works to train OpenAI’s large language models (LLMs) and creating infringing works in the outputs of OpenAI’s consumer-facing AI product, ChatGPT.

ChatGPT relies on OpenAI’s LLMs, which are “trained” by supplying the LLMs with large amounts of text, allowing the models to identify relationships between words in the training data. After training, the LLMs can generate responses to user prompts that resemble human-authored text. Plaintiffs claim that OpenAI used datasets containing their copyrighted works to train its LLMs. Plaintiffs allege that, when prompted, ChatGPT can generate accurate summaries of plaintiffs’ books and outlines for potential sequels, which plaintiffs allege are unauthorized and infringing.

In its motion to dismiss, OpenAI argued that because the consolidated complaint does not cite to or attach examples of the allegedly infringing outputs, the court could not properly evaluate whether ChatGPT’s outputs were substantially similar to plaintiffs’ works. OpenAI further argued that the outputs at issue are analogous to the summaries of news articles that the court determined were not substantially similar to copyrighted works in a decision that Judge Sidney Stein had issued in another action in the MDL, New York Times v. Microsoft Corp. (Read our summary of the court’s decision here.)

The court first held that the complaint “squarely alleges” that OpenAI had access to plaintiffs’ works and that ChatGPT’s allegedly infringing outputs are based on plaintiffs’ works, satisfying the “actual copying” element of a prima facie infringement claim.

The court next determined that although plaintiffs did not attach ChatGPT’s allegedly infringing outputs to the consolidated complaint, it would consider the outputs that plaintiffs submitted in opposition to the motion, which the court found were incorporated by reference into and integral to the complaint. The court excerpted specific examples of the allegedly infringing outputs, such as a detailed summary of the novel A Game of Thrones and an alternative sequel outline, to illustrate how the outputs could be determined to be substantially similar to the original works. The court found that “[a] more discerning observer could easily conclude that [the] detailed summary is substantially similar to [the author’s] original work, including because the summary conveys the overall tone and feel of the original work by parroting the plot, characters, and themes of the original.” The same was true of an excerpted outline for a potential sequel to the same author’s work, which output the court found that a reasonable jury applying the “more discerning observer test” could determine was substantially similar based on the incorporation of copyrightable elements of setting, plot and characters. By contrast, the court noted, the summaries of news articles at issue in New York Times v. Microsoft Corp. merely summarized non-copyrightable elements and differed in style, tone, length and sentence structure from the original works.

The court thus held that “[i]t is sufficient to defeat OpenAI’s motion to dismiss that plaintiffs have alleged some outputs that a reasonable jury could find are substantially similar to plaintiffs’ works.” In so concluding, the court specifically noted that nothing in its decision “is intended to suggest a view on whether the allegedly infringing outputs are protected as fair uses of the original works.”

Summary prepared by Safia Gray Hussain and Chloe Gordils