Trump Admin Splinters Over AI’s Right To Train On Americans’ Copyrighted Work

The Trump administration appears not to have reached a conclusion on whether artificial intelligence should be able to learn from Americans’ work for free.

The Department of Justice (DOJ) told a federal judge that AI training on copyrighted books is fair use, and that licensing would only help big labs and “old mainstream media,” according to a Tuesday court filing. That assertion cuts against the White House’s March AI framework, which said that it is still contested whether that training infringes on copyright.

“It is not in the public’s interest for the largest technology companies to have an oligopoly on LLM [Large Language Model] training due to licensing entry barriers that function primarily as large subsidies for old mainstream media companies,” the DOJ filing read. “[T]he United States has a strong interest in this Court rejecting any argument that training LLMs on copyrighted texts violates copyright law.”

The department’s argued OpenAI copies the works to train its models, but not to republish them.

“An OpenAI LLM thus uses the copyrighted work not to duplicate the work’s expressive content, but as part of a process to learn and act on statistical patterns in written text, including vocabulary, syntax, and knowledge,” the DOJ explained. “The purpose of the copying … differs in kind from the purpose of the copied work.”

The DOJ referred the DCNF to a Sept. 2 X post from associate attorney general Stanley E. Woodward Jr., when asked for comment.

“[President Trump] has made clear that AI dominance is critical to promote national security, prosperity, and economic mobility for all Americans,” Woodward said in the post. “This Administration will never let our Nation be at a disadvantage relative to our foreign adversaries based on a plainly incorrect understanding of copyright law.”

The White House and OpenAI each did not respond to the Daily Caller News Foundation’s request for comment.

One tech policy expert disagrees with the DOJ’s statement.

“The administration believes that AI training is fair use … but they also recognize the complexity of the litigation, the complexity of the issue … And ultimately, they said the court should decide,” Evan Swarztrauber, principal at CorePoint Strategies and a former adviser to FCC Chairs Ajit Pai and Brendan Carr, told the DCNF. “[T]hat’s very different than what the DOJ submitted, which is this fairly aggressive position that nearly all cases of training are fair use, and that licensing is negative. They basically attack licensing.”

Swarztrauber added that the DOJ is wrong to treat licensing as a gift to “old mainstream media.”

“The New York Times is uniquely situated. They have a lot of money to adjudicate their rights. Smaller players don’t,” Swarztrauder told the DCNF. He argued that licensing is the only viable way to compensate these smaller creators.

The New York Times sued OpenAI in 2023 over claims that the AI company’s ChatGPT models were being trained with millions of the outlet’s articles, the lawsuit stated. The NYT argued that these models were journalism competitors because they could “generate output that recites Times content, verbatim, closely summarizes it, and mimics its expressive style.”

Microsoft was also part of the lawsuit, as the NYT alleged Copilot, or Bing Chat at the time, featured verbatim sections of the outlet’s articles in chatbot responses, according to the lawsuit.

Anthropic was similarly accused of building AI models with copyrighted booked.

Anthropic’s Project Panama, uncovered in an Aug. 2024 lawsuit against Claude’s creator, involved purchasing millions of books to then cut off their spines and scan the pages before throwing the books away. The process of destroying these books is depicted in a July 29 video posted on X.