U.S. Department of Justice Intervenes in New York Times v. OpenAI Case, Training Large Language Models on Copyrighted Works Generally Constitutes Fair Use
The U.S. Department of Justice submitted a statement of interest to the Manhattan federal court, supporting OpenAI's claim in the copyright lawsuit filed by The New York Times and other publishers: training large language models on copyrighted works generally constitutes fair use.
Deputy Attorney General Stanley Woodward and Assistant Attorney General Brett Shumate stated in the document that broadly considering training as infringement contradicts the fundamental principles of copyright law, hinders scientific and creative progress, undermines U.S. prosperity and economic mobility, and gives foreign competitors a competitive advantage. Woodward described this as a "historic" statement of interest, emphasizing that the dominance of artificial intelligence is crucial for national security, and the government will not allow a misunderstanding of copyright law to put the nation at a disadvantage. The DOJ argues that the training process is transformative use, aimed at learning statistical patterns in text rather than copying expressions, and that training should be viewed separately from chat outputs; it also stated that the benefits of AI far outweigh competitive harms.
The Times sued OpenAI and Microsoft in 2023, accusing them of scraping and copying reports to train their models. The case has been consolidated with lawsuits from other newspapers and authors, including John Grisham and George R. R. Martin, into a multi-district procedure. Both parties are scheduled to submit motions for summary judgment this week. Times spokesperson Graham James stated that the government is siding with a few trillion-dollar AI companies at the expense of creators whose works have been "stolen"; he also stated that both sides can coexist, and companies should pay for content under copyright law. OpenAI did not immediately comment. The DOJ is not a party to the lawsuit, and its statement is persuasive to the judge but not automatically binding.
This is the first time the federal government has systematically expressed its position in such training data lawsuits. Licensing negotiations and fair use defenses will proceed in parallel: licensed media will pursue authorization, while unlicensed media will go through the courts.
The national security rhetoric connects copyright's four factors analysis to industrial policy, but the court must still evaluate transformative use, market substitution, and the amount used according to statutory law. The individual cases of copying reports at the output end are not dismissed by this document.
In market mechanisms, buyers are model companies that want to acquire corpus with zero licensing costs, while sellers are publishers and author groups that want to label corpus as chargeable assets. The driving force is the government writing training rights into competitiveness, which is a policy event. Funding is being redistributed between licensing fees, litigation reserves, and computational power investments. Beneficiaries are training vendors that have not yet completed full procurement and defendants claiming fair use, while those under pressure are news organizations relying on reprints and subscriptions, as well as content providers treating licensing fees as revenue lines.
Source: Public Information
ABAB AI Insight
The four factors of copyright have been debated for two hundred years, and this time the DOJ has rewritten the fourth factor "market impact" as "national market impact." The Times wants payment per article, while OpenAI argues that statistical learning does not count as copying. The government intervenes not to write code for ChatGPT, but to prevent a situation where training is deemed infringing, making U.S. corpus costs higher than those of competitors who do not adhere to the same rulings. Woodward links "misunderstanding copyright" with "foreign competitors," effectively elevating the Southern District Court's memorandum to the level of industrial warfare. The statement of interest does not guarantee victory, but provides the judge with a political risk explanation.
The capital path is shifting training data from gray area scraping to either fair use or a national licensing market. Microsoft, as a cloud and funding partner, is involved in the same case. Publishers that have signed contracts with OpenAI have effectively locked in prices, while those who have not are betting on the ruling. The motivation is to execute the 2025 AI leadership executive order; the strategy is to insert official explanations into the case file before the summary judgment. The Authors Guild and genre authors are consolidated in the case, indicating that the corpus war covers both news and fiction.
Comparable cases include the transformative use in the Google Books scanning case, the fair use of Sony's recording devices, and the compulsory licensing of music sampling. This time, the focus is on whether the "talking index" replaces the original market. The phase is shifting from corporate defense to sovereign endorsement, where copyright holders must contend not only with the defendants' lawyers.
This signifies a regulatory change. The mechanism of change is that fair use, originally a defense, is being written as an industrial policy tool. Once training and output are accepted by the court as separable, copying occurring in weights does not equate to copying occurring in answers. If publishers lose on the training end, they can only shift to arguing substantial similarity and product substitution on the output end. Pricing power shifts from copyright registration certificates to national strategic documents for models.
ABAB News · Cognitive Law
Once training is framed as national security, copyright becomes an attachment to industrial policy.
Statistical patterns and sentence-by-sentence copying; the court seeks boundaries, not slogans.
The government submits not a ruling, but a risk explanation for the judge's desk.