Context: Last February, The Authors Guild and at least 28 different authors filed a class action against OpenAI in the Southern District of New York, alleging that it infringed plaintiffs’ copyrights in their books by downloading and reproducing plaintiffs’ works, by using those reproduced works to train OpenAI’s AI large language models, and by creating infringing works in the outputs of OpenAI’s LLM products, including ChatGPT (February 5, 2024 class action complaint). The suit was later consolidated with several other similar cases against OpenAI and Microsoft in a multidistrict litigation, and refiled in the same court in June. OpenAI has since moved to dismiss the complaint based on both the output and downloading arguments, because it “fails to plausibly allege substantial similarity between plaintiffs’ works and ChatGPT’s outputs or to cite or attach examples of allegedly infringing outputs for the Court to consider when evaluating the issue of substantial similarity”.
What’s new:
- Judge Sidney H. Stein of the United States District Court for the Southern District of New York has handed down two opinions in the case, with the first denying OpenAI’s motion to dismiss the “output” copyright infringement claim. He notes that the plaintiffs’ allegations satisfy the elements of a prima facie claim of infringement as to at least some outputs of ChatGPT (October 27, 2025 Southern District of New York decision).
- In the second opinion, Judge Stein has allowed the plaintiffs’ theory of infringement based on downloading copies (from so-called shadow libraries) as a separate claim of infringement apart from training and rejected Microsoft’s motion to strike allegations related to GPT-4o and GPT-4o Mini. However, OpenAI did succeed in having Judge Stein strike new allegations related to new models of OpenAI: “GPT-4V, GPT-4.5, GPT-5”, and “derivatives” and “successors” of any model (these were not part of the prior complaints).
Direct impact: According to Professor Edward Lee of ChatGPTIsEatingTheWorld, this decision means things have gone “from bad to worse” for OpenAI (28 October, 2025 ChatGPTIsEatingTheWorld article). The class plaintiffs now have at least three different theories of infringement, meaning the company really has its work cut out for it, Mr. Lee writes. Before this case goes to trial, there will be a summary judgment decision on fair use. The case is potentially turning into Bartz v. Anthropic II or even Bartz II-plus, he believes. Bartz is a copyright class action in which the plaintiffs have struck a $1.5 billion settlement with Claude AI maker Anthropic (August 26, 2025 ai fray article).
Wider ramifications: This development comes in addition to the fact that of the current 57 copyright lawsuits against AI companies in the U.S. (October 27, 2025 ChatGPTIsEatingTheWorld article), 12 involve OpenAI (all as a defendant). The company also appears to be on the losing track in its suit against German music rights collecting society Gesellschaft für musikalische Aufführungs- und mechanische Vervielfältigungsrechte (GEMA) (September 29, 2025 ai fray article), and faces copyright lawsuits in several other jurisdictions, including India (January 27, 2025 ai fray article) and Canada (December 3, 2024 ai fray article). However, the GEMA trial in Germany last month shows that non-U.S. jurisdictions are perhaps a lot stricter because foreign equivalents of the fair use defense are much narrower exceptions than fair use in the U.S.
Mr. Lee believes that, with the way the class action is proceeding, this case could turn into a second occurrence of Bartz v. Anthropic. In this case, and in many other AI copyright infringement cases, there are three key battlefields:
- how the AI companies originally obtained the material (pirate libraries, or so-called shadow libraries, in this case), which Judge Stein has allowed as an argument;
- LLM training, which is where the defense argument of fair use comes in; and
- output, which Judge Stein has also allowed.
Fair use was a key argument of Anthropic’s in Bartz, which it succeeded on in the Northern District of California. In June, the court granted a summary judgment for Anthropic that the training use of the copyrighted books and the print-to-digital format change were both “fair use” (June 24, 2025 ai fray article). This decision could have been appealed if the case hadn’t been settled.
While Judge Stein has allowed output as an argument for now, it may be more of an alternative route for the plaintiffs. If there is a pirate library issue here, then this argument is more likely to help them succeed, especially after the Anthropic settlement.
In Bartz v. Anthropic, the claimants focused on the first two lines of argument: so input, not output. Output is indeed excluded from the settlement, so if it turns out that authors can additionally recover something for output, then all those authors (individually or in one or more new classes) could get even more for their works.
Mr. Lee says if Judge Stein follows Judge William H. Alsup’s approach in Bartz v. Anthropic, the case may indeed turn into “Bartz II”:
“Even if OpenAI wins its defense for fair use for AI training, it still must win on the Shadow Library claim by convincing Judge Stein that the correct approach is to treat the initial download as part of the purpose of training.”
While all three questions are new for courts, judges seem unwilling to accept the first one (pirate libraries), unless plaintiffs present a weak case.
According to Justin Nelson of Susman Godfrey, counsel to the plaintiffs in this case and in Bartz v. Anthropic (whom ai fray recently interviewed: October 28, 2025 ai fray article), piracy is not new and has been around since before the Internet. Some pirate websites said they were dying (and even going out of business) before AI, he told ai fray, and AI has “reinvigorated” them.
But, he also noted in the interview, Bartz sends the signal that piracy is wrong:
“We cannot ignore how this data was acquired: strip-mining humanity’s creations without compensation.”
When it comes to fair use, Mr. Nelson also noted that many cases are still in preliminary and discovery stages, but so far, the balance of decisions and the expert judgment of the U.S. Copyright Office is against them:
“I’m hopeful that, at the end of the day, courts will see the huge adverse effect that these AI models are causing to creators and copyright holders.”