A brief with teeth, if not binding force
On September 1, 2026, the Department of Justice filed a Statement of Interest in In re: OpenAI, Inc. Copyright Infringement Litigation — the sprawling multidistrict case before Judge Sidney H. Stein in the Southern District of New York that consolidates The New York Times' suit and related claims from other publishers and authors against OpenAI and Microsoft. A Statement of Interest doesn't bind the court; it's the executive branch telling a judge what outcome it thinks the law — and the national interest — demands. DOJ's answer here is unambiguous: training large language models on copyrighted text is fair use, full stop, and it's asking Judge Stein to say so (CourtListener MDL docket).
The brief calls LLM training "exceedingly transformative" and takes direct aim at Kadrey v. Meta Platforms, the Northern District of California case in which Judge Vince Chhabria granted Meta summary judgment on fair use for 13 named plaintiffs on June 25, 2025. DOJ isn't objecting to the outcome in Kadrey — Meta won — it's objecting to the reasoning. Chhabria found training highly transformative but held that the fourth fair-use factor, market harm, could still doom AI companies if plaintiffs proved "market dilution": that LLMs let anyone flood a market with competing works, cheapening the originals. The plaintiffs in Kadrey failed to build that record, so Meta won narrowly. But Chhabria left the theory alive for the next set of plaintiffs — plaintiffs that now include The New York Times (Goodwin Procter analysis). DOJ's brief is an attempt to kill that theory before it reaches OpenAI.
The steelman: market dilution is a real copyright harm
Before dismissing DOJ's critics, it's worth taking the market-dilution theory seriously, because it isn't a fringe argument — it's the position the U.S. Copyright Office itself took. In its May 2025 Part 3 report on generative AI training, the Office explicitly rejected the idea that training is automatically or "exceedingly" transformative, calling it "at best, modestly transformative" when outputs compete with the works they were trained on. It went further than Chhabria, treating market dilution — AI-generated content flooding a genre or style and depressing the value of human-authored originals — as a legitimate, government-endorsed harm theory (Copyright and Artificial Intelligence, U.S. Copyright Office). The Times spent real money reporting stories that a model can now paraphrase at scale for free; that's precisely the kind of substitutive harm copyright's fourth factor was built to catch. DOJ's own brief is, in effect, arguing against a position its sister agency already published.
That's an unusual split for the federal government to litigate against itself, even informally, and it undercuts DOJ's framing of "exceedingly transformative" as settled consensus rather than one contested reading among several.
Where DOJ's argument holds up
Still, the market-concentration point deserves more credit than it's gotten from critics. DOJ warns that a licensing mandate would let only the largest incumbents — OpenAI, Google, Microsoft, Meta — absorb the compliance and litigation costs, freezing out smaller labs and open-source developers. The Copyright Office's own report backs this concern indirectly: it recommended letting voluntary licensing markets develop but conceded that one-off deals "may not scale." A litigation-driven damages regime, decided case by case in front of different judges applying an inherently mushy four-factor test, is a worse mechanism for building a scalable market than a statutory framework would be. If Congress wants creators compensated without entrenching the four best-funded AI labs, a collective-licensing structure — closer to how ASCAP and BMI clear music rights — is a far more coherent tool than amicus briefs asking courts to declare a blanket rule.
The credibility problem DOJ created for itself
What weakens DOJ's brief isn't its legal argument — it's the appearance of self-dealing around it. Above the Law reported that the administration has been in talks to take a roughly 5% equity stake in OpenAI, a position CNN valued near $42.6 billion, discussions that were reportedly underway well before this filing (Above the Law). The brief never discloses that. Whatever the legal merits of DOJ's fair-use position — and there are real merits — a Justice Department that may become a financial stakeholder in the defendant should say so when it files a brief arguing for that defendant's core legal theory. Omitting it hands ammunition to exactly the critics DOJ should want to disarm.
A US win wouldn't be a global win
Even total victory for OpenAI in Judge Stein's courtroom wouldn't resolve the company's copyright exposure. The EU AI Act's Article 53 requires general-purpose AI providers operating in EU markets to respect rightsholders' text-and-data-mining opt-outs under the bloc's 2019 Copyright Directive, regardless of where training took place (TNW). A favorable US fair-use precedent settles the domestic fight; it does nothing for the parallel European one.
Our take
DOJ is right that a per-case, per-judge fair-use lottery is a bad way to build a licensing market, and right that the compliance costs of one could entrench today's largest labs. But "exceedingly transformative" overstates a genuinely contested legal question, contradicts the Copyright Office's own findings, and the undisclosed equity talks make the intervention look less like principled policy and more like the government picking winners it may soon co-own. The fix is legislative: a scalable, statutory licensing mechanism that compensates creators without making case outcomes as unpredictable — or as politically compromised — as this brief now makes them look.