Fate of AI and Fair Use Copyright Report Will Depend on Judges

Federal courts—not the US Copyright Office—will have the final say on implementing any recommendations from the office’s forthcoming report on the fair use ramifications of training AI on protected works.

The first part of the report, released in July, called on Congress to pass a law protecting people from unauthorized deepfakes. Register of Copyrights Shira Perlmutter recently told senators the office plans to publish by the end of 2024 a framework for analyzing the fair use doctrine’s applicability to AI’s use of copyrighted works.

Judges across the country are presiding over roughly three dozen lawsuits against OpenAI Inc., Microsoft Corp., Meta Platforms Inc., Anthropic PBC, and other AI companies. How they apply that framework as they wrestle with novel legal questions remains to be seen—and it may be Congress that benefits the most from the Copyright Office’s analysis.

The Copyright Act spells out the four factors judges must consider in a fair use analysis, but the precise contours have been defined through court precedent. The Copyright Office—an arm of Congress—can’t manage how judges approach fair use, Harvard University law professor Rebecca Tushnet said.

“The courts have made it pretty clear that the Copyright Office does not have authority to interpret infringement law,” Tushnet said.

The office’s role in compiling public feedback will produce analysis quicker than the pace of litigation, Santa Clara University law professor Edward Lee said.

“We just have this rather unique situation of AI litigation proceeding on one track—and with no decision yet—and the AI study on another track by the Copyright Office,” Lee said.

Perlmutter, the office’s director, told lawmakers in October the office plans to release the two remaining sections of its report this year.

Lee said he doesn’t expect the office to take a categorical stance on fair use. Instead, it’ll likely produce a framework for how to think about the debate that “a court could look at it as almost akin to an amicus brief,” he said.

Even if courts don’t fully adopt the report’s approach to what’s ultimately a jury question, they may find it useful in unpacking how generative AI actually works, said Aaron Moss, a partner at Greenberg Glusker Fields Claman & Machtinger LLP.

“This is the first time that we’re collecting all of this information from all the various stakeholders,” Moss said. “If they do a good job in synthesizing that information—as I have no doubt that they will—it will at the very least serve as a road map for courts and the public just to better understand how these models actually work.”

Stakeholder Input

The office received more than 10,000 public comments to its October 2023 notice of inquiry into the legal issues raised by AI, according to its website.

Stakeholders—including AI companies, creatives, engineers, and legal professionals—were encouraged to give their opinions on when, or if, the unauthorized use of copyrighted works to train AI models constitutes fair use.

The doctrine’s four factors probe the purpose and transformativeness of the use of the copyrighted work, the expressive nature of the original, the portion used, and the potential market impacts of the new use.

OpenAI argued training an AI model represents a transformative, permissible use of copyrighted works—echoing its arguments in ongoing litigation. The American Society of Composers, Authors and Publishers told the office AI leads to illegal reproductions that merely replace the original works.

Sen. Marsha Blackburn (R-Tenn.) said in her letter that unlicensed AI ingestion of copyrighted works shouldn’t be fair use when the output “competes commercially” with human-created media.

The forthcoming report will likely synthesize the comments and provide policymakers and judges with an outline of the debates’ contours, Moss said.

“We are keenly aware of widespread interest in our AI and copyright initiative, including from the courts,” the office said in a statement to Bloomberg Law, adding that it embarks on studies “from the unique position we occupy.”

“We cannot predict how our Report will be used, but we know courts, litigants, and others have relied on our studies in the past,” it said

Congressional Action

The office’s analysis may ultimately be “more helpful to Congress than it is going to be for the courts,” Brown Rudnick LLP partner Michael Graif said.

While public comments suggested consensus that existing federal copyright law is adequate to deal with AI debates, Perlmutter conceded to a US Senate Judiciary subcommittee that lawmakers may be pressured to step in.

“Once court decisions start being issued, we may start to see parties come to Congress asking for some change,” Perlmutter said at the Nov. 13 hearing.

If courts “leave openings” for the unauthorized use of commercial writing for AI training purposes, “we will need legislation in order to preserve copyright incentives for writers,” the Authors Guild said in its letter.

OpenAI’s letter urged the office to “proceed cautiously” in calling for new legislation that may be “premature or misguided” as the technology evolves. Courts haven’t had an opportunity yet to “pass on most of the questions” raised in the office’s notice of inquiry, it said.

Graif said he expects the AI fair use question will ultimately reach the US Supreme Court. But Congress may also be motivated to act—and that decision, he said, would be “highly informed” by the office’s report.

Lawmakers are “going to defer to the Copyright Office’s clear expertise in the area in deciding whether to intervene legislatively,” Graif said.

While Tushnet, the Harvard professor, agreed the report might be more useful to lawmakers, she said it may take an industry-backed deal to pass anything concrete.

In drafting copyright law, lawmakers wanted to avoid the need to come back and amend legislation every time a new technology emerged, Lee said.

“Instead of that, fair use does a lot of the work,” Lee said. “That leaves it to courts to figure out whether it’s infringement or fair use.”