IP Law Daily, AI NEWS: U.S. government supports fair use determination in OpenAI copyright litigation, (Sep 2, 2026)

By Thomas Long, J.D.

In a statement to court in a consolidated class-action infringement suit, the government highlights the transformative nature of using copyrighted works to train LLMs and argues that the training has no negative market effects.

As the parties in the consolidated class-action litigation against Open AI, Inc., asserting copyright infringement in the training of OpenAI’s large-language models (LLMs), prepare their summary judgment arguments, the United States government has filed a statement informing the court of its interest in the matter. Asserting that effective integration of artificial intelligence technology will have critical national security implications, the government warns against “rules of law that make it significantly more difficult to develop a robust AI industry in the United States.” The statement argues that OpenAI’s conduct is protected by copyright fair use principles, largely because of the “exceedingly transformative” nature of the training of LLMs on written works. The government also takes aim at the novel “market dilution” theory embraced by at least one court in its fair-use analysis, contending that “a wholesale rewriting of copyright principles” is best left to elected representatives rather than the courts (In re OpenAI, Inc., Copyright Infringement Litigation, No. 1:25-cv-03483-SHS-OTW (S.D.N.Y. Sep. 1, 2026)).

National security implications. The statement starts by reiterating the government’s position that the U.S. has a strong interest in promoting “a robust and competitive artificial intelligence industry that sets the standard for the practice and procedure of AI use globally,” as set forth in a January 2025 executive order issued by President Trump. The government asserts that it “thus has a strong interest in the question whether training AI models on written works constitutes ‘fair use’ under copyright law.” Excessive restrictions on AI development will “threaten national security and give a competitive advantage to foreign adversaries who are not so encumbered,” according to the government.

Fair-use principles. Pointing out that the constitutional basis for intellectual property protection is the promotion of innovation, the government states that the training of AI models on copyrighted material does not violate copyright laws. The plaintiffs’ attempt to narrow fair-use law, the government argues, would impede progress. The statement notes that one key plaintiff—The New York Times—has begun using LLMs for journalistic purposes, and that others have used the technology to achieve scientific breakthroughs. “Constraining LLM development under a misunderstanding of fair use doctrine would thwart such creative and scientific progress while hindering American prosperity and economic mobility,” the government says.

With respect to arguments that copyright concerns stemming from AI training should be addressed through thought licensing, the government takes the position that this approach risked hampering competition in the market for LLMs. “[O]nly the largest technology companies might have the capital necessary to pay licensing fees,” argues the government, “And such licensing fees would disproportionately benefit legacy media outlets due to the sheer volume of their written publications.”

Transformative use. With respect to the first of the four fair-use factors listed in 17 U.S.C. § 107—the “purpose and character of the use”—the statement borrows language from a federal district court in calling the copying of articles for LLM training “spectacularly transformative.” Bartz v. Anthropic PBC, 787 F. Supp. 3d 1007, 1021 (N.D. Cal. 2025). “An OpenAI LLM … uses the copyrighted work not to duplicate the work’s expressive content, but as part of a process to learn and act on statistical patterns in written text, including vocabulary, syntax, and knowledge,” the government opines.

“To be sure, at the output (rather than training) stage, certain uses may not be transformative if the LLM reconstructs and disseminates an original copyrighted work,” the government concedes, but it argues that this should not bear on the transformative nature of training uses.

Market effects. Jumping to the fourth fair use factor—“the effect of the use upon the potential market for or value of the copyrighted work”—the government tells the court that the transformative purpose of the AI training use can be overcome only if the resulting LLM provides text in a manner that offers a competing substitute for the plaintiffs’ works. Copies used to train LLMs do not serve as substitutes for the originals and do not diminish the protected market opportunities for them, according to the government. The statement notes that training does not itself display or reveal any of the training materials to the public; it only uses a copy to “teach” the LLM about relationships between data. “The potential for future outputs that might cause market harm is simply not relevant to evaluating an LLM training use under the required use-by-use analysis,” the government argues. The statement also takes the position that potentially competing outputs still would not infringe unless they were substantially similar to the copyrighted works at issue.

The statement levels criticism at an “indirect substitution” theory of “market dilution” adopted by a district court in Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026, 1052-57 (N.D. Cal. 2025). In that case, the court took a position on the market effects factor that some commentators have called “striking.” The market dilution theory is based on the potential “use [of] LLMs to create books and then sell them, competing with books written by human authors.” Id. at 1052, 1054. In the government’s view, the Kadrey court’s position is “deeply flawed” because it collapses LLM training and LLM outputs into a single continuous use. The court then “applied a capacious, genre-level understanding of ‘substitution’” while failing to recognize that training was its own distinct use that does not involve any substitution. Moreover, the government contends that any LLM outputs that might compete with copyrighted works used in the training process are unlikely to infringe due to a lack of substantial similarity.

The statement also mentions in a footnote that Register of Copyrights Shira Perlmutter, in the third part of the Copyright Office’s report on Copyright and AI (focused on generative AI training), “appeared to endorse a similar theory.” However, the government says that “[Perlmutter’s] understanding does not warrant deference,” and the statement criticizes the Register’s “threadbare reasoning.”

Conclusion on fair use. While conceding that fair use is ordinarily a fact-specific inquiry, the statement argues that it would be problematic and legally incorrect to impose broad copyright liability that would hinder LLM training. That training, the government says, is consistent with the creative progress that is copyright’s fundamental constitutional objective.

“At a minimum, the creative possibilities and public benefits that LLM training advances far outweigh any competitive harm (even assuming such harm is cognizable),” the government asserts. “The entire reason that AI models are rapidly reshaping the economy and national security is that they help people, including those working in creative fields, make things and get things done.”

The case is No. 1:25-cv-03483-SHS-OTW.

Judge: Stein, S.

Attorneys: Joseph Richard Saveri (Saveri Law Firm, LLP) for Sarah Silverman. Andrew Michael Gass (Latham & Watkins LLP) for OpenAI Inc. Michael Weisbuch, Senior Counsel to the Associate Attorney General, for the United States of America.

Companies: OpenAI Inc.

MainStory: TopStory AINews Copyright TechnologyInternet NewYorkNews GCNNews