Attempted Firing of Register of Copyrights Puts Future AI Protections at Risk

June 1, 2025

Ben Kessler  -  Ben Kessler, AFM Director of Government Affairs

Article I, Section 8, Clause 8 of the US Constitution grants Congress the authority to, “promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries.” Interference with that enumerated power, like the president’s attempted dismissal of Register of Copyrights Shira Permutter, should worry everyone.

I write “attempted” because such action is clearly outside the president’s authority. And it should worry working musicians the most, regardless of their political affiliation. To provide context, I will walk through how we got here.

On May 8, President Trump fired Librarian of Congress Carla Hayden. Dr. Hayden was in the ninth year of her 10-year term. She was also the first professional librarian to serve as Librarian of Congress in decades. To be clear, while the Library itself is part of the legislative branch of government, the president does appoint the Librarian. But still, the seemingly sudden firing raised questions with just one year left on the decade-long term. This is relevant because the US Copyright Office is within the Library of Congress, and the Librarian of Congress appoints the head of the Copyright Office, the Register of Copyrights.

The Copyright Office’s Copyright and Artificial Intelligence reports series appears to be the reason for the attempt to fire Perlmutter. The Office published Part 1 on digital replicas or deepfakes in July 2024. It then published Part 2 on the copyrightability of generative artificial intelligence (AI) in January of this year. At the time of Hayden’s firing, the Office had not released Part 3 on generative AI and the legal implication of machine learning.

On May 9, less than 24 hours after Hayden was removed from office, the Copyright Office released a “prepublication” version of Part 3 “in response to congressional inquiries and expressions of interest from stakeholders.” Additionally, it stated, “a final version of Part 3 will be published in the near future, without any substantive changes expected in the analysis or conclusions.”

True to its mission to serve copyright owners and users, the report does not establish any brightline rules. For better or worse, those are rarely, if ever, found in copyright law. It does, however, make some important assertions and observations, some of which I have outlined here.

For additional context, many from tech are asserting that all generative AI is fair use. Fair use is an affirmative defense to copyright infringement. In short, if all machine learning is a fair use, musicians have no ability to stop AI developers from using their work for machine learning.

The four fair use factors are:

1. the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
2. the nature of the copyrighted work;
3. the amount and substantiality of the portion used in relation to the copyrighted work as a whole;
4. the effect of the use upon the potential market for or value of the copyrighted work.

As a shorthand, the first factor, based on case law, refers to whether the new work is “transformative.” In its report, the Office rejected two common arguments about the transformative nature of AI training. The first argument is that the use of copyrighted works is not for expressive purposes, i.e., it’s the output, not the input, that matters. The Office responded to that claim stating, “Where the resulting model is used to generate expressive content, or potentially reproduce copyrighted expression [such as a song] the training use cannot be fairly characterized as ‘nonexpressive.’”

The report went on to include, that the Office does not “… agree that AI training is inherently transformative because it is like human learning.” In rejecting the claim, the Office encapsulated perfectly what it is to be an author, in copyright terms: “Humans retain only imperfect impressions of the works they have experienced, filtered through their own unique personalities, histories, memories, and worldviews.”

Some of the most consequential parts of the report for musicians deal with that fourth factor: market harm. The Office states, “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.” And the potential market harm is not limited to outputs, but inputs as well. “Market harm can also stem from AI models’ generation of material stylistically similar to works in their training data.” And while copyright law does not protect style, per se, “stylistic imitation made possible by its use in training may impact the creator’s market.”

At approximately 6 p.m., on May 9, the Copyright Office posted Part 3 of its report. By 9 p.m. on Saturday, May 10, news reports confirmed that President Trump had attempted to fire Register Perlmutter. Given the timeline, the reason for the termination notice appears to be clear. The Copyright Office, under Register Perlmutter’s leadership, faithfully summarized its findings based on well-established legal principles. Unfortunately, it seems, anything less than subservience to tech interests was a fireable offense.

---