US DOJ backs fair use for AI training in NYT copyright suit against OpenAI
The US Department of Justice has filed a brief in the consolidated copyright lawsuit brought by the New York Times against OpenAI and Microsoft, siding with the AI companies and arguing that training AI models on copyrighted material qualifies as fair use.
The Times sued OpenAI and Microsoft in late 2023 in a Manhattan federal court, alleging that millions of its articles were used without permission to train models such as GPT-4 and to build products that compete with the paper as an information source. The Times cited billions of dollars in damages and demanded the destruction of language models trained on its articles. The case has escalated considerably since then and is widely seen as a bellwether for how courts will treat copyright and AI training generally.
The DOJ's filing argues that the copyrighted text used for LLM training in this case does not amount to infringement, because training and output are legally distinct. During training, entire works are copied but never made publicly available, and the DOJ says the resulting outputs 'often if not always lack substantial similarity' to the originals; a blanket theory of market harm that conflates training with output is, in the department's view, legally wrong.
To illustrate the point, the filing invokes Joan Didion: as a teenager she copied Hemingway's stories to learn how his sentences worked. The DOJ argues that under the logic of the Kadrey ruling, Didion could have faced liability whenever she later published, since her learning process and subsequent writing would have been treated as a single use. Citing an earlier ruling, the department says it would be unthinkable to require people to pay whenever they later draw on a book to write something new in a new way. The DOJ adds that LLMs carry creative and public value, that human beings create original works using them, and that liability for AI training would stifle the creativity copyright law is meant to protect.
The filing also directly rebuts a US Copyright Office report, authored under then Register Shira Perlmutter, that had rejected blanket fair use for AI training; that report argued AI works with perfect copies and generates content at a speed and scale far beyond human creation, and that commercial applications competing with original works in existing markets exceed what fair use allows. The DOJ says Perlmutter's assessment carries no binding legal authority and that the report ignored case law on case-by-case analysis and on the types of market harm that count under the statute.
Perlmutter was fired by the Trump administration shortly after the report came out. Democratic Representative Joe Morelle said she was dismissed because she refused to legitimize AI training on copyrighted works, a position he said Trump ally Elon Musk favored. The DOJ's filing notes in a footnote that Perlmutter is currently challenging her dismissal. No court ruling or case outcome has been reported; the filing states the department's argument, not a decision.
Key facts
- The US Department of Justice filed a brief in the consolidated New York Times copyright suit, siding with OpenAI and Microsoft and arguing that training AI models on copyrighted material is fair use.
- The New York Times sued OpenAI and Microsoft in late 2023 in a Manhattan federal court, citing billions of dollars in damages and demanding the destruction of models trained on its articles.
- The DOJ argues training and output are legally distinct: the resulting outputs 'often if not always lack substantial similarity' to the copyrighted originals used in training.
- The filing directly rebuts a US Copyright Office report by former Register Shira Perlmutter, who was fired by the Trump administration shortly after publishing it and is now challenging her dismissal.
- The DOJ uses a Joan Didion analogy, arguing that under the Kadrey ruling's logic, her teenage practice of copying Hemingway's stories to learn his style could have made her liable whenever she later published.
Why it matters
A brief from the US Department of Justice is a rare, high-weight intervention in a private lawsuit: it puts the federal government's legal position behind the AI industry's core defense in a case widely treated as a bellwether for how courts will handle copyright and AI training across the board. The filing also fits a broader pattern, cited in the same report, of the Trump administration taking a pro-AI stance in copyright fights, including its rejection of the Copyright Office's earlier report against blanket fair use.
Who it affects
OpenAI and Microsoft, the defendants who stand to benefit from the DOJ's position; the New York Times and other rights holders in the consolidated suit, whose claim for damages and model destruction the brief argues against; and, more broadly, any company training models on copyrighted text and any publisher or author whose work is used that way. Shira Perlmutter, the fired Copyright Office official whose report the DOJ's filing rebuts, is directly named and is separately contesting her dismissal.
How to use it
This is a legal filing, not a ruling: there is nothing to license or adopt yet. AI companies can point to the DOJ's position as a favorable government argument in ongoing and future disputes, but it carries no binding legal authority on its own and does not resolve the New York Times case. Publishers and authors tracking the litigation should treat this as one side's brief within an active suit, not as a decided outcome.
How solid is it
The report draws on the DOJ's actual court filing, quoting it directly, and lays out the chain of the New York Times lawsuit and the Copyright Office report it responds to. No date is given for when the DOJ filed the brief, only the New York Times' original late-2023 filing date, and no court ruling or case outcome is reported: the story covers only the DOJ's filed argument, which remains unresolved.
Risks and caveats
The DOJ's position is one side's legal argument, not settled law, and the report notes plainly that others disagree. The DOJ's own filing has to work around the scale objection the Copyright Office raised: a single author copying text to learn is different from a multibillion-dollar company turning that content into competing mass-market products. Perlmutter's firing is itself politically contested, with a Democratic lawmaker attributing it to her refusal to endorse blanket fair use for AI training, a characterization the DOJ's filing does not itself confirm.
“Human beings create original works using LLMs.”
— US Department of Justice, in its court filing