•
The most closely watched IP dispute of the generative-AI era will decide whether unlicensed model training is fair use. Inside the consolidated In re: OpenAI proceeding and its landmark discovery fights.
In re: OpenAI, Inc. Copyright Infringement Litigation — U.S. District Court for the Southern District of New York (Hon. Sidney H. Stein; Mag. J. Ona T. Wang)
The most closely watched intellectual-property dispute of the generative-AI era remains unresolved, and its trajectory will likely determine the economics of model training for a decade. The Times filed suit in December 2023, alleging that OpenAI and its investor Microsoft ingested millions of its articles to train large language models without license or compensation, and that the resulting systems can regurgitate protected expression verbatim. The case has since absorbed related actions by other publishers and authors into a consolidated proceeding before Judge Stein, making it the central vehicle through which the federal courts will confront whether unlicensed training constitutes fair use under 17 U.S.C. § 107.
Procedurally, the litigation cleared its first major hurdle in the spring of 2025, when Judge Stein declined to dismiss the core infringement and contributory-infringement claims while paring back several ancillary theories. That ruling preserved the plaintiffs' central contention—that copying at the scale of model training is not excused by the transformative-use doctrine—and set the parties on a discovery path of extraordinary breadth. No trial date has been set, and the fair-use question that animates the case has not yet been decided on the merits.
Discovery itself has produced some of the most significant rulings to date. The plaintiffs sought production of a large sample of ChatGPT conversation logs to test how often the model reproduces copyrighted text, and Magistrate Judge Wang ordered production of a twenty-million-log sample over OpenAI's objections. In January 2026, Judge Stein affirmed that order, rejecting the argument that the magistrate's decision was clearly erroneous and finding that user-privacy interests were adequately protected by the discovery safeguards in place. The dispute over user logs previews the evidentiary fights to come, in which the frequency and fidelity of memorized output will bear directly on both infringement and the fourth fair-use factor.
The stakes are difficult to overstate. Statutory damages for willful infringement run to $150,000 per work, and the sheer number of asserted registrations means that an adverse verdict could translate into liability of a magnitude no defendant has faced in a copyright case. More important than any single award is the precedential question: whether the transformative-use logic that protected search indexing and snippet display extends to systems that learn from, and can reproduce, the expression they were trained on.
For practitioners, the case is a live laboratory for questions that will recur across every AI docket—how to define the relevant "use," how to measure market harm to a licensing market that is itself nascent, and how discovery obligations attach to training corpora and model weights. However the fair-use question is ultimately resolved, the record being built in the Southern District of New York will frame the analysis for every generative-AI dispute that follows.
Judge Stein largely denied the motions to dismiss on April 4, 2025, letting the core infringement and contributory infringement claims proceed while trimming ancillary theories. He also rejected the argument that claims tied to the training of earlier models were time-barred.
The discovery rulings have been more consequential than the pleading ruling. Magistrate Judge Wang ordered production of a twenty-million-log sample of ChatGPT conversations, and in the January 5, 2026 order affirming the discovery rulings, Judge Stein affirmed that order over OpenAI's objection, finding the privacy interests adequately protected by the safeguards in place. Summary judgment briefing was scheduled to conclude in April 2026.
Fair use has not been decided. Anyone telling you this case has resolved whether training is fair use is describing something that has not happened.
The April 2025 opinion on the motions to dismiss frames every claim that survives. The January 2026 order on the log production explains the court's reasoning about privacy against evidentiary need, which is the template other AI cases will follow. And the summary judgment briefing, once public, contains the parties' full fair-use arguments in a form no article reproduces.
Pull the current filings here: In re OpenAI copyright litigation. Consolidated proceedings generate heavy filing weeks around briefing deadlines, so case alerts will catch a ruling faster than a manual check.
The image-generation version of the same fight, and the one closest to trial, is Andersen v. Stability AI. For the Supreme Court's 2026 statement on when a service provider answers for infringement by its users, read Cox Communications v. Sony Music. And for a dispute over AI systems accessing another company's property rather than its content, see Amazon v. Perplexity.
Read the available docket record. These pages reflect the records collected so far; check the source for subsequent filings.
District Court, S.D. New York · Filed December 27, 2023