This is a 20-page statement of interest filed by the United States Department of Justice on September 1, 2026 in the consolidated multidistrict litigation In re: OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143 (SHS) (OTW) in the Southern District of New York, docketed at 1:25-cv-03483-SHS-OTW as Document 316 and marked "This Document Relates To: All Matters." It is signed by Associate Attorney General Stanley E. Woodward, Jr., Assistant Attorney General for the Civil Division Brett Shumate, and Senior Counsel to the Associate Attorney General Michael Weisbuch.
The filing argues that copying copyrighted text in order to train a large language model is fair use under 17 U.S.C. § 107; that training and output are distinct uses requiring separate fair-use analysis; and that the market-dilution reasoning in *Kadrey v. Meta Platforms* misapplies the fourth factor by collapsing the two. It is the position of the executive branch, filed as a non-party, and not a holding.
Posture and stated interest
The United States appears under 28 U.S.C. § 517, which permits the Department of Justice to "attend to the interests of the United States in a suit pending in a court of the United States." A footnote states that the statute "contains no time limitation and does not require the Court's leave," citing Gil v. Winn Dixie Stores, Inc., 242 F. Supp. 3d 1315, 1317 (S.D. Fla. 2017). Another footnote disclaims any contention that the conduct at issue was authorized by, consented to by, or for the benefit of the government under 28 U.S.C. § 1498. Although the brief refers for simplicity only to the New York Times and OpenAI, it states that its arguments "apply similarly to all parties in this litigation and the related cases, including book authors and publishers."
The asserted interest rests on three executive-branch instruments. Executive Order 14179, "Removing Barriers to American Leadership in Artificial Intelligence" (January 23, 2025), is quoted for the policy "to sustain and enhance America's global AI dominance in order to promote human flourishing, economic competitiveness, and national security." The executive order of June 2, 2026, "Promoting Advanced Artificial Intelligence Innovation and Security," is quoted for a commitment to "encouraging AI innovation and accelerating responsible AI adoption across government and industry" and for the proposition that advanced AI "introduce[s] new national security considerations." The Legislative Recommendations accompanying the National Policy Framework for Artificial Intelligence (March 2026) are quoted for the balance that "American creators, publishers, and innovators should be protected from AI-generated outputs that infringe their protected content, without undermining lawful innovation and free expression," and for the statement that the "training of AI models on copyrighted material," in itself, "does not violate copyright laws." The brief also cites a 2022 Government Accountability Office publication on AI and national security, and argues that "[r]ules of law that make it significantly more difficult to develop a robust AI industry in the United States therefore threaten national security and give a competitive advantage to foreign adversaries."
The competition and licensing argument
One passage inverts the framing usually advanced by AI developers' critics. The brief argues that an adverse fair-use ruling "would hamper competition in the market for LLMs, because only the largest technology companies might have the capital necessary to pay licensing fees," and that licensing fees "would disproportionately benefit legacy media outlets due to the sheer volume of their written publications." It states that "[i]t is not in the public's interest for the largest technology companies to have an oligopoly on LLM training due to licensing entry barriers that function primarily as large subsidies for old mainstream media companies."
A footnote qualifies the argument in several respects: the United States "takes no position on whether a licensing regime would be financially or logistically feasible given the breadth of works at issue, the liability that developers would face, and coordination difficulties"; it says that whether or not paid licensing "would pose an existential threat to the American AI industry, it would at least hamper innovation and put American AI companies at a competitive disadvantage relative to competitors located in other countries that do not consistently respect U.S. intellectual property law"; and it notes that publishers "could enter (and have entered) into licensing agreements to provide developers with specialized access to real-time, pay-walled, proprietary, and other content" regardless of how the fair-use question is resolved.
The brief argues that LLMs "can and should help level the playing field between mainstream and independent publishers," because authors with limited resources can use them to compete — its example is generating an image for an article that "otherwise might require a photographer or license" — and because LLMs "can direct users to dissenting sources."
The staged model of training
The brief adopts the court's own description of LLM development from New York Times Co. v. Microsoft Corp., No. 1:23-cv-11195-SHS-OTW, ECF No. 514 (S.D.N.Y. April 4, 2025): a developer first collects and stores data at the acquisition or pre-training stage; then trains by making copies of the works and converting them into numeric representations from which the model learns linguistic patterns; then, at the output stage, responds to user queries subject to guardrails and filters. "Each stage may present distinct questions of copyright law," and the United States addresses only the training stage.
Fair-use framework
The legal section recites the Intellectual Property Clause and the principle from Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340, 349 (1991), that "[t]he primary objective of copyright is not to reward the labor of authors" but to promote progress, together with Twentieth Century Music Corp. v. Aiken, 422 U.S. 151, 156 (1975), for the instruction that "[w]hen technological change has rendered its literal terms ambiguous, the Copyright Act must be construed in light of this basic purpose." It quotes Judge Leval's opinion in Authors Guild v. Google, Inc., 804 F.3d 202, 212 (2d Cir. 2015), and through it Cary v. Kearsley, 170 Eng. Rep. 679, 681 (1802) ("[O]ne must not put manacles upon science"), for the proposition that absolute control over copying would limit rather than expand public knowledge.
On the statutory structure, it notes that the § 107 preamble's six uses are illustrative rather than exhaustive, and quotes the House Report for the intent that "there is no disposition to freeze the doctrine in the statute, especially during a period of rapid technological change." It records the shift in emphasis among the factors: Harper & Row Publishers v. Nation Enterprises, 471 U.S. 539, 566 (1985), once called the fourth factor "undoubtedly the single most important," while after Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994), the Second Circuit observed that the Court "stressed also the importance of the first factor."
The organizing principle of the brief is the use-by-use requirement drawn from Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 533 (2023): "[t]he same copying may be fair when used for one purpose but not another." It marshals Sony Corp. of America v. Universal City Studios, Inc., 464 U.S. 417 (1984), Campbell, Google LLC v. Oracle America, Inc., 593 U.S. 1 (2021), and the separate treatment of search, snippet and library-download uses in Authors Guild as instances of the same discipline.
The first factor
The brief argues that training copying is "transformative—spectacularly so," borrowing the phrase from *Bartz v. Anthropic PBC*, 787 F. Supp. 3d 1007, 1021 (N.D. Cal. 2025), and "highly transformative" from Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026, 1044 (N.D. Cal. 2025). The reasoning is that an 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," so that "[t]he purpose of the copying (to build an intelligent, interactive model) differs in kind from the purpose of the copied work (to use language to directly entertain or educate a reading audience)." Google v. Oracle is offered as the closest analogue: Google copied code "precisely" but for the distinct purpose of letting programmers "put their basic skills to use in a new one."
The brief concedes that output-stage uses "may not be transformative if the LLM reconstructs and disseminates an original copyrighted work," but argues that this "should not bear on the transformative nature (or any other aspect) of the training use." A footnote adds that "any prospective remedy for infringement would need to be limited to what is necessary to afford plaintiffs complete relief," citing Trump v. CASA, Inc., 606 U.S. 831, 852–54 (2025), and that "[a] tiny sliver of anomalous reconstructive outputs would not support a remedy that cuts off or threatens massive liability" for training uses. Commerciality is dismissed on Campbell's observation that "[n]early all of the illustrative uses listed in the preamble paragraph of § 107 . . . are generally conducted for profit."
A further footnote disposes of the second and third factors. The second is said to follow the transformative purpose. On the third, although training "involves the copying of entire works at an early step," what matters under Authors Guild "is not so much the amount and substantiality of the portion used in making a copy, but rather the amount and substantiality of what is thereby made accessible to a public for which it may serve as a competing substitute" — and training "does not make any copied works accessible to the public."
The fourth factor and the critique of Kadrey
The fourth-factor argument turns on substitution. Drawing on Campbell, Harper & Row and Authors Guild, the brief argues that what counts is "significant substitutive competition," not competition generally, and that the question is whether copying is "done in a manner that results in widespread revelation of sufficiently significant portions of the original as to make available a significantly competing substitute." Its central factual claim is that "training does not reveal anything to the public at all," so "[t]he 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."
It extends the point to outputs: those that "copy no protected elements of the original work" cannot cause cognizable harm merely by falling in "the same genre or category of works," since "a genre is an uncopyrightable idea or method of expression." The supporting authorities are Edward Lee, Copyright Dilution Under Constitutional Scrutiny, 25 Chi.-Kent J. Intell. Prop. 1 (2026), Peters v. West, 692 F.3d 629, 636 (7th Cir. 2012), and Abdin v. CBS Broadcasting Inc., 971 F.3d 57, 70 (2d Cir. 2020).
The brief then addresses Kadrey directly, calling its market-dilution discussion "[c]ontrary dicta in a Northern District of California decision" that "misapplies copyright principles to LLM training." It records that the Kadrey court adopted the indirect-substitution theory "[w]ithout the benefit of briefing," noted that "the plaintiffs barely g[a]ve this issue lip service," and appeared to acknowledge that the theory has "never made a difference in a case before." The asserted error is that the court "improperly collapsed LLM training and LLM outputs into a single continuous use, then applied a capacious, genre-level understanding of 'substitution,'" thereby overlooking "arguably the most basic element of copyright infringement—whether the 'works are substantially similar.'"
The illustration offered is Joan Didion, who as a teenager typed out Ernest Hemingway's stories "to learn how the sentences worked" (Linda Kuehl, Joan Didion, the Art of Fiction No. 71, The Paris Review, Issue 74, Fall–Winter 1978). "By the Kadrey court's logic," the brief argues, "Didion should have incurred liability to Hemingway every time she published a piece," quoting Bartz that "to make anyone pay specifically for the use of a book . . . each time they later draw upon it when writing new things in new ways would be unthinkable."
A footnote addresses the Copyright Office. It states that the Kadrey court "did not cite any case supporting such a theory," but that the Register of Copyrights, "who is currently challenging her removal, appeared to endorse a similar theory" in Copyright and Artificial Intelligence Part 3: Generative AI Training (May 2025). The brief asserts that "[h]er understanding does not warrant deference," citing Loper Bright Enterprises v. Raimondo, 603 U.S. 369, 412 (2024), and describes the Register's reasoning as "threadbare."
The section closes on public benefits, which Google permits the fourth factor to weigh. It argues that training's creative and scientific benefits "far outweigh any competitive harm (even assuming such harm is cognizable)," noting that New York Times authors themselves use LLMs to "conceptualize and edit" articles (Futurism, May 12, 2026), and citing an independent writer's LLM-assisted critique of data-center water-usage reporting (Andy Masley, January 13, 2025; Pirate Wires, December 18, 2025). It adds that "[w]hether this new technology warrants a wholesale rewriting of copyright principles is a policy question best left to the People and their elected representatives."
Relation to the underlying rulings
The filing defends training on the basis that training reveals nothing to the public; it says nothing about how training material was acquired. That distinction matters against Bartz, where Judge Alsup held training "spectacularly" transformative but treated the retention of pirated copies in a central library separately, and where the case settled on that ground rather than on the training holding (Bartz v. Anthropic). The brief quotes only the transformative-use passages.
Its characterization of Kadrey as dicta reached "[w]ithout the benefit of briefing" matches that decision's own posture: Judge Chhabria ruled for Meta because the plaintiffs "failed to present meaningful evidence" on market effect, stated the ruling did not establish that Meta's use was lawful, and raised market dilution as a theory the plaintiffs had not adequately pursued but that could favor future plaintiffs (Kadrey v. Meta). The United States and the Kadrey court agree on the procedural posture and disagree on whether the theory is sound.
Reception
The filing surprised the U.S. Patent and Trademark Office and the Copyright Office, according to sources cited by Axios on September 19, 2026, which reported that the position had not been coordinated among the bodies that work closely on the issue and that no career antitrust attorneys signed it alongside the senior officials, unlike many statements of interest. Publishers criticized its claims, including the argument that enforcing copyright is too cumbersome and would threaten US AI leadership over foreign rivals; Ziff Davis chief executive Vivek Shah argued in an op-ed that the reasoning was muddled, and Axios noted the gap between the brief's description of licensing enforcement as cumbersome and its footnote declining to take a position on whether a licensing regime is feasible (Source: axios.com).
Provenance
Retrieved from CourtListener's RECAP archive at storage.courtlistener.com. Corroborated against Reuters, "US government backs OpenAI in New York Times copyright case" (September 2, 2026); Wired, "Trump Administration Sides With OpenAI in New York Times Lawsuit" (September 2, 2026), which links the same document; and the filing's own ECF stamp (Case 1:25-cv-03483-SHS-OTW, Document 316, Filed 09/01/26) and signature block dated September 1, 2026.
Relationships
- supports: NYT v. Microsoft, OpenAI et al., OpenAI
- contradicts: Kadrey v. Meta
- depends-on: AI Copyright, Executive Order 14179 — Removing Barriers to American Leadership in Artificial Intelligence, EO — Promoting Advanced AI Innovation and Security (Trump, signed June 2, 2026)
- related: Bartz v. Anthropic, Authors Guild v. OpenAI, Microsoft, United States Copyright Office