The dust is still settling in the aftermath of the Copyright Office’s issuance of a “pre-publication” version of its long-awaited Report analyzing the issue of fair use in AI training. A cloud now hangs over the Report–and whether it’s DOA–following President Trump’s firing of the Register of Copyrights Shira Perlmutter less than 24 hours after its posting.
It’s possible this Report never becomes official, especially not if the Trump Administration disagrees with its analysis–a distinct possibility given Trump’s Executive Order on Removing Barriers to American Leadership in AI.
Let me offer two reasons why I think the U.S. government should not adopt this pre-publication report: one reason relating to institutions and the statutory role the Copyright Office plays, and the second relating to the internal inconsistency of the Report’s analytical approach.
The Statutory Role of the US Copyright Office in Advising Congress
The Register of Copyrights invoked her “statutory responsibility [in Section 701(b)(1), (b)(4)] to ‘[c]onduct studies’ and ‘[a]dvise Congress on national and international issues relating to copyright’” as the basis for issuing its Report.
Advising Congress is one thing. But making rulings on each factor of fair use based in part on comments submitted from some of the parties on both sides in the 36 copyright lawsuits being litigated in courts is quite another.
Indeed, I cannot think of any novel legal issue related to fair use in the pending 36 copyright lawsuits–from transformative purpose in AI training, use of pirated books datasets, deployment of AI in RAG search, and a new and untested theory of market dilution–that the Copyright Office didn’t rule on.
And by “rule on,” I mean the Office took a legal position, favoring one side or even party in the litigation.
For example, embarking on admittedly “uncharted territory” (p. 65), the Copyright Office endorsed a completely new and untested theory of “market dilution” under Factor 4 that is so expansive it seeks to protect a copyrighted work from competition in genres or types of works even if they are noninfringing. If an AI model was trained on romance novels, the copyright holder gets to stop the AI’s capability from generating new, noninfringing romance novels because, in the Office’s view the new works might dilute the market for the original novels and “more competition for sales of an author’s works and more difficulty for audiences in finding them.” The Copyright Office concluded the same with apparently all AI-generated music–and even relied on the comments from UMG, which is a plaintiff in 2 pending copyright lawsuits against Udio and Suno. Protecting copyright holders from competition from noninfringing works in entire genres turns the Copyright Clause on its head. Instead of promoting progress, the goal is to protect copyrights.
I’ll say more on the substance of this new theory of market harm below. But I think it exposes the flaw in the Copyright Office’s unbounded scope in the Report. This isn’t advice to Congress any more. This is a pseudo-amicus brief ruling on a specific new theory raised by an interested party in the copyright litigation. Indeed, the Office even quotes directly from UMG’s complaint (in footnote 375, p. 65).
Section 701 of the Copyright Act does not authorize the Copyright Office to hold its own shadow proceeding ruling on legal issues being raised by parties in active litigation. To do so usurps the province of the federal courts, especially given that fair use is a quintessentially judge-made doctrine. It’s for the courts, not the Copyright Office, to embark on “uncharted territory” in fair use. If the Office wanted to give its opinions on live controversies involving fair use, the proper procedure for the Office to use is filing an amicus brief in the pending litigation, which comports with due process and gives the opponent a chance to respond to the Office’s views. Indeed, the Copyright Office’s Report is itself unprecedented. To my knowledge, the Office has not ever issued a Report taking the side of one party in active litigation on novel issues of copyright law.
The Copyright Office’s Report Is Contradictory in Its Analysis and Approach
Now, let’s turn to substance. Parts of the Report’s analysis I agree with. Other parts, I do not. I did not expect to agree with 100% of the Report’s analysis, so I will save my quibbling over specific issues for another forum (for a preview, see my initial analysis of the market dilution theory). But the part seems most objectionable is the apparent double standard the Report uses.
Category 1: For some issues related to fair use, the Report rejected views that “go[] beyond established fair use boundaries.”
Category 2: But, for other issues related to fair use, the Report felt free to embark on “uncharted territory” and endorse views of fair use without any established legal precedent, such as the untested theory of market dilution and copyright protection from competition for entire genres (e.g., AI-generated romance novels and music).
Category 1 and 2 broke along party-lines. Positions favorable to defendant AI companies fall within Category 1–the Copyright Office rejected them. Positions favorable to plaintiff copyright holders fall within Category 2–the Copyright Office accepted them, even crediting UMG.
If the Copyright Office stuck to the “established fair use boundaries,” it should not have embarked on “uncharted territory” to endorse a new theory floated by an interested party in the ongoing litigation. Alternatively, if the Office felt free to go beyond the established fair use boundaries into uncharted territory, it should apply the same standard throughout its analysis. Be consistent.
Related Stories