The fight over AI training and copyright has moved from a technical legal dispute into a broader clash over creative markets, government authority, and the future rules for generative AI. A pre-publication report from the US Copyright Office rejected the sweeping claim that all AI training should be treated as fair use, and the fallout was immediate.
A day after the report appeared, the Trump administration fired Shira Perlmutter, the head of the Copyright Office. That timing fueled speculation that the report helped trigger her removal, especially because the guidance challenged a central argument made by artificial intelligence firms.
The Report Put Limits Around AI Fair Use
The Copyright Office did not say that AI training is never fair use. Its position was more nuanced: some training may qualify, but not every use of copyrighted work becomes fair just because it supports an AI system.
The office emphasized that courts are the institutions that ultimately decide fair use. Still, its report outlined how courts may weigh the issue, especially when an AI model competes with the same creative markets that supplied its training material.
The report followed the Copyright Office's review of more than 10,000 comments about whether creators should be compensated when their works are used in AI training, and whether such compensation is practical. The office framed the issue as a need to balance public interest in “maintaining a thriving creative community” with public interest in “allowing technological innovation to flourish.”
Two fair use factors received particular attention: the character of the use, including whether it is transformative, and the effect on the market. The office indicated that those two questions are likely to carry significant weight in court.
Where Training May Be More Defensible
The Copyright Office suggested that AI training is more likely to be fair use when the resulting model does not compete with the original creative works. It gave examples of models serving a different function, including language models that translate texts, moderate content, or correct grammar.
For audio, the report drew a distinction between technology that helps producers clean up unwanted distortion and technology that generates songs in the style of popular artists. The first kind of use may be more defensible because it solves a different problem. The second raises sharper market concerns because it can substitute for work associated with creators.
The office also acknowledged that “training a generative AI foundation model on a large and diverse dataset will often be transformative.” But it warned that “not every transformative use is a fair one.” If a model's function serves the same purpose as the copyrighted works used to train it, the fair use argument becomes harder.
That concern applies to situations such as chatbots regurgitating news articles, as alleged in The New York Times' dispute with OpenAI over ChatGPT. In those cases, the office said, “unless the original work itself is being targeted for comment or parody, it is hard to see the use as transformative.”
The report also pointed to filters as one possible way AI companies could preserve chatbot usefulness while reducing infringement risk. The office referred to tools that could “prevent the generation of infringing content.”
Why Creators And AI Firms Disagree
The Copyright Office described the stakes as high because the same technology can support innovation and threaten creative livelihoods. Creator-side commenters warned about the consequences of unlicensed training. The office summarized their concern by saying, “Commenters painted a dire picture of what unlicensed training would mean for artists’ livelihoods.”
Industry advocates made the opposite warning. They argued that giving artists the power to hamper or “kill” AI development could produce “far less competition, far less innovation, and very likely the loss of the United States’ position as the leader in global AI development.”
The report did not fully endorse either extreme. Instead, it recognized that the legal result depends on the specific use, the model's purpose, and the degree of market harm. It also encouraged more licensing as a way to compensate creators.
Courtney Radsch, the director of the Center for Journalism & Liberty at the Open Markets Institute, said the report raised issues that powerful technology interests did not want recognized. According to Radsch, the Copyright Office identified that it remains an open question how much data AI developers need to build effective models, that a consent framework may be needed beyond forcing creators to opt out, and that “AI trained on copyrighted works could replace original creators in the marketplace.”
The Firing Turned A Legal Debate Political
Perlmutter's firing came after Donald Trump removed Librarian of Congress Carla Hayden. NPR noted that Hayden was the first African American to hold that post. The Library of Congress houses the Copyright Office, which made the removals especially contentious for critics who said these institutions should operate independently of the executive branch.
White House Press Secretary Karoline Leavitt said Hayden's removal was tied to “quite concerning things that she had done at the Library of Congress in the pursuit of DEI and putting inappropriate books in the library for children.”
Rep. Joe Morelle (D.-N.Y.) condemned Perlmutter's removal as “a brazen, unprecedented power grab with no legal basis.” He argued that Trump was trampling Congress' authority and suggested that Musk and other technology leaders trying to dominate AI could benefit from intervention at the Copyright Office.
Morelle connected the timing directly to the AI report. “It is surely no coincidence [Trump] acted less than a day after she refused to rubber-stamp Elon Musk’s efforts to mine troves of copyrighted works to train AI models,” he said, apparently referring to Musk's xAI chatbot, Grok.
Radsch also linked the firing to the report, saying Perlmutter's removal “appears directly linked to her office’s new AI report questioning unlimited harvesting of copyrighted materials.” She described the episode as an executive intrusion into the Library of Congress after months of lobbying by corporate billionaires.
What Happens Next
The report released on Friday was not finalized, but the source article said it was not expected to change radically unless Trump's new acting head intervened to overhaul the guidance. That possibility added another layer of uncertainty to an already unsettled legal area.
Tensions appeared to continue after the firing. Social media posts on Monday described an apparent standoff at the Copyright Office between Capitol Police and men rumored to be with Elon Musk's Department of Government Efficiency (DOGE). A source familiar with the matter told Wired that the men were “Brian Nieves, who claimed he was the new deputy librarian, and Paul Perkins, who said he was the new acting director of the Copyright Office, as well as acting Registrar,” while Wired reported that it remained “unclear whether the men accurately identified themselves.”
A Capitol Police spokesperson told Wired that no one was escorted off the premises or denied entry to the office.
The legal question remains in the courts. In a case involving Meta's torrenting of a pirated books dataset to train AI models, US District Judge Vince Chhabria said he did not immediately “understand how that can be fair use.” He added, “You have companies using copyright-protected material to create a product that is capable of producing an infinite number of competing products.”
For AI companies, the message is that fair use cannot be assumed across the board. For creators, the report gives official weight to concerns about consent, licensing, and market substitution. For everyone else, it shows that the rules governing AI training are still being contested at the intersection of copyright law, technology policy, and institutional power.