The report released late Friday — a day before the firing of the head of the Copyright Office — is the third and final installment in a series from the government agency analyzing questions at the crossroads of copyright law and the rapid rise of AI technologies. The reports' conclusions have been drawn from the more than 10,000 comments the office received from various stakeholders in late 2023, including from AI developers, individuals and industry groups that represent content creators.
The government agency's latest report emphasizes that using copyrighted material for AI training may sometimes be highly transformative and allowed by fair use, but some commercial uses might not be covered if the models produce outputs that supplant the original works.
"On one end of the spectrum, training a model is most transformative when the purpose is to deploy it for research, or in a closed system that constrains it to a non-substitutive task," the Copyright Office said. "For example, training a language model on a large collection of data, including social media posts, articles, and books, for deployment in systems used for content moderation does not have the same educational purpose as those papers and books."
At the other end of the spectrum, according to the report, are models that generate outputs that are substantially similar to the training content.
"For example, a foundation image model might be further trained on images from a popular animated series and deployed to generate images of characters from that series," the report said.
The report comes as courts face the question of fair use in dozens of copyright lawsuits against AI developers, including a proposed class action from a group of bestselling authors against Meta Platforms, where a California federal judge is considering summary judgment motions from each side.
The government agency's three reports on AI were never meant to be binding on lawmakers or the courts, so it's unclear what impact Saturday's firing of Shira Perlmutter as director of the Copyright Office will have on how its conclusions are received. The Trump administration did not provide a reason for why Perlmutter was fired.
Still, the office's report underscores the complexity of fair use confronting courts and policymakers as AI technologies develop at breakneck speed, often by relying on copyrighted content for training.
The fair use analysis has also been made more challenging by the U.S. Supreme Court's 2023 decision in Andy Warhol Foundation for the Visual Arts v. Goldsmith
, which the office said "clarified the concept of transformativeness" — which weighs heavily in favor of fair use because it means a work has been used to create something new and for a different purpose than the original.In Warhol, justices held that a silkscreen of music icon Prince that Warhol created based on a photograph by Lynn Goldsmith was not fair use, regardless of how transformative, because both works shared the same commercial purpose of magazine publishing
"The court explained that while adding new expression can be relevant to evaluating whether a use has a different purpose and character, it does not necessarily make the use transformative," the Copyright Office said. "Even significant alterations will not be enough if the use ultimately serves a purpose similar to that of the original, and may instead produce a derivative work and demonstrate the 'need for licensing.'"
The Copyright Office said the high court has twice described the fourth and final fair use factor — market impact — as the most important element in the fair use analysis, although its weight will vary depending on the strength of other factors. The government agency cautioned against reading the fourth factor too narrowly.
"The statute on its face encompasses any 'effect' upon the potential market. The speed and scale at which AI systems generate content pose a serious risk of diluting markets for works of the same kind as in their training data. That means more competition for sales of an author's works and more difficulty for audiences in finding them," the report said. "If thousands of AI-generated romance novels are put on the market, fewer of the human-authored romance novels that the AI was trained on are likely to be sold."
The report recommends minimal government intervention, stressing that licensing markets should continue to develop voluntarily and saying that the flexibility of the fair use doctrine has historically allowed copyright law to adapt to new technologies, even if they present fresh challenges, such as AI.
Congress and stakeholders have been waiting for several months for the third and final report, at one point prompting a letter from lawmakers. Friday's report is a "pre-publication version" released in response to congressional inquiries and interest from stakeholders, according to the Copyright Office, saying "a final version will be published in the near future, without any substantive changes expected in the analysis or conclusions."
The other reports from the Copyright Office on AI that were based on the thousands of comments it received in 2023 focused on digital replicas, or deepfakes, and on whether outputs from generative AI systems qualify for copyright protection.
--Editing by Dave Trumbore.
For a reprint of this article, please contact reprints@law360.com.
