DOJ backs OpenAI’s fair-use defense in New York Times copyright fight
The DOJ backed OpenAI and Microsoft on September 1, arguing that training large language models on copyrighted texts should not violate copyright law. Seven House Democrats demanded records.
By Grace Kim
4 min read

What's News
- The DOJ filed a statement of interest on September 1 supporting OpenAI and Microsoft in the New York Times’ copyright suit.
- The Times and other publishers allege OpenAI and Microsoft used millions of articles without permission to train large language models.
- Rep. Chellie Pingree and six other House Democrats requested DOJ draft documents and a list of outside companies.
- A June 2025 ruling favored Anthropic on book training; a September 2026 ruling rejected fair use for copied Westlaw summaries.
- The bipartisan CLEAR Act, S. 3813, would require disclosure of copyrighted works in AI training datasets.
By Grace Kim
The Department of Justice filed a statement of interest on September 1 supporting OpenAI and Microsoft in The New York Times’ copyright lawsuit. The filing placed the government behind a broad fair-use defense for training large language models on copyrighted material.
The Times and other publishers allege that OpenAI and Microsoft used millions of articles without permission to train large language models. OpenAI and Microsoft say the training was fair use.
What did the DOJ tell the court?
A Congressional Research Service legal sidebar quoting the filing said the DOJ told the court:
“the United States has a strong interest in . . . rejecting any argument that training [large language models] on copyrighted texts violates copyright law.”
That position goes beyond the White House’s National Policy Framework for Artificial Intelligence, published in March 2026. The framework says the Trump administration believes training on copyrighted material does not violate copyright law. It also acknowledges contrary arguments and supports letting courts resolve the issue.
The Carolina Principles for Emerging Technologies, adopted by G20 ministers at the Commerce Department’s Innovation Ministerial, take a more cautious approach. They recognize complex legal and policy questions and the essential role of copyright protections in supporting creators.
The DOJ has chosen a more categorical position than either the White House or the G20.
Why are House Democrats demanding records?
A letter circulated this week by Rep. Chellie Pingree (D-Maine) and six other House Democrats said:
“the intervention stands to undermine public trust in the impartiality of the DOJ and its ability to represent the best interests of the nation.”
They were alarmed that the Copyright Office and the U.S. Patent and Trademark Office were not consulted before the filing. The lawmakers described the agencies as legally required to provide impartial input on copyright matters.
The letter asks for all draft documents and a list of any outside companies that influenced the DOJ’s position.
Axios reported that no career antitrust attorneys signed the September filing alongside senior DOJ officials. That was a departure from the typical practice for statements of interest. The Justice Department has not responded to requests for comment.
House Oversight Committee incoming chair Robert Garcia has sent document-preservation letters to Trump administration officials, including Attorney General Todd Blanche. The letters lay groundwork for investigations as Democrats are poised to take control of both chambers in January.
The Pingree letter opens what could become extensive congressional oversight.
How have courts treated AI training?
Courts have delivered mixed results, with outcomes turning on the type of system, the works copied and the harm shown.
- June 2025: A California federal judge held that Anthropic’s training on books was fair use.
- In a separate case in the same district, another judge ruled for Meta on similar facts. The judge warned that factor four, market harm, could carry decisive weight against fair use in other cases.
- September 2026: The Third Circuit ruled against fair use for a non-generative AI system that copied Westlaw case summaries.
The Copyright Office’s 2025 pre-publication report concluded that some uses of copyrighted works for generative-AI training will qualify as fair use and others will not. The rulings leave no single rule for all AI training.
What could change?
The bipartisan CLEAR Act (S. 3813), introduced by Sens. Adam Schiff (D-Calif.) and John Curtis (R-Utah), would require AI developers to disclose copyrighted works in training datasets. The bill would not resolve whether the uses are fair.
The business stakes center on the cost of licensing. The DOJ’s filing argues that licensing costs would burden small AI developers and threaten U.S. AI leadership and national security.
Critics including former USPTO deputy director Coke Morgan Stewart, writing in Fox News, say large AI firms are better able to pay. They argue licensing markets are already emerging. Several AI startups cited by Fox News say paying for training data reduces legal uncertainty and improves their models.
The statement of interest is not binding, but it carries weight with the court and gives OpenAI and Microsoft the government’s support. The next tests are judicial treatment of licensing and market harm, along with any congressional action in January. Those decisions will set the cost of training data and the economics of generative AI for years to come.
Source: epyhia.com
More from Grace Kim
Show full bio
Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,582 articles