- Lead. The US Department of Justice filed a brief on September 2 supporting OpenAI and Microsoft in the New York Times copyright lawsuit, arguing that training large language models on copyrighted text generally qualifies as fair use under existing law.
- Fact. The filing marks the first time the US government has formally intervened in AI copyright litigation, invoking national security and economic competitiveness as grounds for protecting AI training practices.
- Stake. If the DOJ position holds sway with the presiding judge, it would effectively remove a foundational legal risk facing every major AI developer — while leaving publishers and authors to seek redress through Congress rather than the courts.
What the Government Argued
The Justice Department’s filing, described in multiple accounts of the brief, made several interlocking claims. Training LLMs on copyrighted text is transformative in the legal sense — it produces a statistical model rather than reproducing the underlying works — and therefore falls within the traditional scope of fair use. The DOJ further warned that restricting this practice would “severely hamper tech progress” and harm America’s position in a technology competition with China, where domestic AI developers operate under no equivalent legal constraint.
The national security framing was the most politically charged element. Lawyers for the government stated explicitly that “the success of America’s artificial intelligence industry is an important national security interest” — language that aligns the administration’s position on AI training with its broader industrial policy rationale for protecting US technology champions from litigation-driven fragmentation.
The NYT Case and the Broader MDL
The New York Times sued OpenAI and Microsoft in December 2023, alleging that the companies reproduced its journalism in training data without authorisation, causing measurable commercial harm. The case was subsequently folded into a multi-district litigation umbrella that now includes suits from book authors, news publishers, and creative professionals �� making the government’s intervention particularly consequential, as any ruling on the fair use question would shape outcomes across that entire docket.
The Times reacted sharply. Its legal team criticised the DOJ for “backing AI companies” and argued that the government’s national security framing was a post-hoc rationalisation to shield commercially powerful defendants from accountability. The filing landed as a direct rebuke to a coalition of media organisations that had lobbied the administration throughout 2025 for a licensing-based approach to AI training data.
Implications for the Industry
The DOJ’s intervention sets up a clear judicial test: courts must now weigh a formal government position that AI training is transformative fair use against the plaintiffs’ claim that it represents wholesale, uncompensated copying. Legal experts noted that the fair use analysis is fact-specific and no single government brief is binding on the court, but the filing substantially raises the evidentiary weight on OpenAI’s side of the scale.
For the broader AI sector, the stakes extend well beyond OpenAI. The case has direct implications for every foundation model developer that has used web-crawled or licensed data from news sources, books, or code repositories. A ruling against the fair use position would introduce licensing costs — and retroactive liability exposure — that analysts estimate could run to tens of billions of dollars across the industry. The DOJ’s filing signals that at least the current administration would treat that outcome as contrary to the national interest.
A parallel proceeding involving Sony and Warner’s lawsuit against Anthropic over music used to train the Claude model is proceeding on a separate track, and it remains to be seen whether the government will file a similar brief there.