AI Copyright Case: Publishers Tell Court to Disregard DOJ Brief

The government picks a side, and the publishers push back

The biggest copyright fight over generative AI in the United States has become a fight about the federal government too. On September 28, The New York Times, the Daily News and other publishers suing OpenAI and Microsoft asked Judge Sidney H. Stein to give the Justice Department's recent brief backing the AI companies "no weight at all."1 In their filing, the outlets said the administration "may have been influenced by extraneous considerations—up to and including the Administration's political and financial interests." They pointed to the AI industry's ties to the White House, including political donations.2

The response targets a statement of interest the Justice Department filed on September 1 in the consolidated multidistrict litigation in Manhattan.11 Some outlets date that filing to September 2, but the substance is the same everywhere.1517 The department called training large language models on written works "exceedingly transformative." It said the market-harm factor "heavily favors fair use." It also warned that broad liability that would generally require licenses for AI training would be "problematic—and legally incorrect."13 According to Reuters, this was the first time the U.S. government has taken a position in litigation over using copyrighted material to train AI.5

The dispute matters well beyond one newspaper's lawsuit. The department said its arguments apply to every coordinated case in MDL 3143, including the class actions brought by book authors.13 The case is also nearing a decision. Cross-motions for summary judgment were filed on September 4, replies are due November 20, and no hearing or ruling date has been set.1314

What the publishers are arguing

Publisher-friendly and neutral outlets describe the response the same way. It makes four main points.

The DOJ filed before seeing the evidence. The publishers say the statement came before the summary-judgment briefs were filed and before much of the record was unsealed. That record includes internal company documents and testimony suggesting executives saw their consumer AI products as substitutes for news subscriptions.1 By their count, the department skipped two years of fact discovery: hundreds of documents, depositions from more than 70 witnesses and dozens of expert reports.14

Policy is not a fair use factor. The publishers say the brief mostly restates the administration's view that unrestricted AI development is a national security priority. They argue that question is separate from the four statutory fair use factors.1 They also cite Supreme Court precedent holding that courts do not give up their judicial role just because the executive branch invokes national security.4

Licensing is affordable. The department suggested content costs could burden new AI companies. The publishers call that a "hypothetical burden on hypothetical entrants." They note that new firms already pay for capital, chips and electricity, and that licensing fees "are barely a footnote" by comparison.1 They point to Microsoft's reported spending of more than $100 billion on OpenAI's development. By the DOJ's logic, they argue, Microsoft should have handed over those resources for free.4

The Copyright Office was ignored. The publishers say the department did not consult the expert agencies. Instead, they argue, it tried to undercut the Copyright Office by calling its reasoning "threadbare" and mentioning the administration's attempt to remove the Register of Copyrights.2 Axios reported that the Patent and Trademark Office and the Copyright Office were caught off guard by the filing, and that no career antitrust attorneys signed it.219

A Times spokesperson said the filing looked like "an effort to further enrich the world's largest companies—and the Administration's largest donors—while undermining this country's free press."2 One trade outlet reported that the publishers also cited reports that the administration was considering an offer of a 5% stake in OpenAI.4 That is the sharpest form of the bias claim. It appears in the publishers' filing as reported, and no court has made any finding on it.

Where the reporting diverges

The coverage agrees on the main events but differs on some framing and background facts. The Hill reported that the publishers asked for "little weight" for the brief, while Bloomberg Law quoted a request for "no weight at all."21 Both phrases may come from the same filing, but they are different requests. The Copyright Office's leadership is also described inconsistently. One account says Shira Perlmutter still runs the office after Trump tried to fire her.4 Another says the office has been without her since she was fired and that her lawsuit over the removal is ongoing.19 Readers should treat the office's status as unsettled. The publishers' broader point, that the agency with expertise was sidelined, does not depend on how that question is resolved.

There are also gaps in the coverage. Pieces that focus on the bias claim tend to leave out one limit on the DOJ brief: it addresses only training, not what the models produce for users.1319 That matters because much of the publishers' evidence of market harm concerns outputs. Their brief cites Microsoft data showing that Copilot sent 83% to 93% fewer clicks to the Times and Daily News websites than traditional Bing search.13 OpenAI points to a different number. Its expert found 24 instances of verbatim reproduction in a sample of 20 million ChatGPT conversation logs.13

The documents that complicate the defense

The timing hurts the government's position. After the DOJ filed, unsealed documents showed that people inside both companies had doubts. A Microsoft applied-science director wrote that the public would soon see models "hoovering up" their work as "an astonishing theft of unprecedented proportions." The same executive wrote that a fair use defense would arguably "make a complete mockery of the idea of 'fair use.'"13 Nick Turley, OpenAI's head of ChatGPT, reportedly wrote that publishers faced an "existential threat" and that the products were "largely substitutive, period."13

Internal opinions do not decide fair use, and one trade outlet noted that the quotes are embarrassing but not necessarily relevant to the legal test.4 Still, they fit the market-substitution theory the publishers are building. The government's brief was written without them.

A courtroom crowded with allies

The dispute has drawn in a wide range of outside groups. By the October 30 amicus deadline, OpenAI had published excerpts from twelve filers supporting fair use for training, including library groups, disability groups and law professors. It accused the Times of pursuing "an undeserved payday."14 On the other side, the News/Media Alliance with ProPublica, Digital Content Next, SAG-AFTRA, the writers guilds and book publishers filed in support of the Times.14

The opposition to the DOJ is not split along party lines. House Judiciary Democrats Jamie Raskin, Jerry Nadler and Hank Johnson filed a brief saying the department had acted contrary to standard practice under this and earlier administrations.9 Conservative groups also filed a brief, warning that letting AI companies "recklessly crawl the internet" was at odds with the administration's own goals, and linking weak copyright protection to Chinese "distillation" attacks.3 Mike Davis of the Article III Project urged the department in a Fox News column to withdraw the filing.16 When allies of the administration and its opponents both object to the same brief, the claim that it reflects a pro-industry tilt rather than settled law gets harder to dismiss.

The appellate signal from legal tech

A ruling in a separate case shifted the debate in the same week the publishers filed. On September 29, the Third Circuit affirmed in Thomson Reuters v. ROSS Intelligence that ROSS's use of Westlaw headnotes to train an AI legal-research tool was not fair use. It is the first federal appellate ruling on fair use in AI training.23 The court found the use only minimally transformative. It also identified harm both to Westlaw's market and to a "rapidly developing" market for licensing headnotes as AI training data.2328

The second point, that a licensing market for training data is real and its loss counts as harm, directly contradicts the DOJ's suggestion that licensing requirements would be legally wrong. Publishers have welcomed it for that reason.26 The ruling has clear limits, though. ROSS's product returned existing text instead of generating new material, and a footnote says the generative-AI arguments in the OpenAI litigation do not apply.24 Fair use advocates criticized the reasoning and hope its reach stays narrow.21 Both sides now have precedent to cite. The DOJ relies on Judge Alsup's finding in Bartz v. Anthropic that training on books was transformative. The publishers rely on Judge Chhabria's view in Kadrey v. Meta that news publishers have "even stronger arguments against fair use."13

Our reading

A statement of interest does not bind the court, and Judge Stein will decide how much weight it gets.13 The publishers have the stronger procedural argument. A brief that does not engage the factual record, the fourth-factor evidence or the expert agency is unlikely to carry much weight in a fact-heavy fair use analysis. The department itself conceded that fair use depends on the facts of each case.4

The bigger effect may be outside this courtroom. The administration has signaled where federal AI and intellectual property policy is headed, and the backlash shows that "AI dominance" does not translate neatly into "no licensing." Litigation keeps spreading. USA Today Co. sued OpenAI and Microsoft on October 8 seeking more than $250 million.17 Canadian publishers filed against Anthropic in Ontario.10 Google is reportedly paying around 100 publishers for their contribution to AI answers.6 The licensing market the DOJ played down is already forming. If Judge Stein rules that training is fair use but outputs that substitute for articles are not, the government's training-only brief could end up mattering much less than the attention around it suggests.