The Fatal Failures of the Copyright Office’s Report on AI
In May 2025, the U.S. Copyright Office released its third report on fair use and artificial intelligence (AI), addressing whether training AI on copyrighted works constitutes fair use. While the report’s main conclusions recognized the strength of existing copyright law and the case-by-case nature of fair use, its reasoning revealed significant flaws that undermine its reliability.
Market Dilution Is Not a Valid Theory of Harm
The most controversial aspect of the report is its endorsement of a new theory called “market dilution.” This theory suggests that an overabundance of creative works—particularly AI-generated ones—can reduce the market value of human-created works. The report argues that any impact on a copyrighted work’s potential market, including competition from new works, weighs against fair use.
This stance misinterprets copyright’s purpose. Copyright protects against unauthorized copying, not competition. Learning from existing works before creating new ones is fundamental to creativity, whether by humans or machines. If copyright protected against market competition in this way, many common creative activities—student writing, amateur photography, self-produced music—would be unfairly targeted.
Legal precedent confirms this. The U.S. Supreme Court has stated that creative works compete even when their audiences overlap and that there is no presumption of market harm from uses that go beyond mere duplication for commercial exploitation. Simply put, copyright law does not shield creators from competition; it protects control over specific uses of their works.
Publishers Do Not Hold Absolute Control Over Access to Their Works
The report also claims that whether copyrighted material was accessed with the owner’s permission affects the fairness of its use. This logic is circular: if permission were obtained, there would be no need to argue fair use.
Rightsholders sometimes argue that training AI on data obtained through “unlawful access” poisons the entire model. This ignores the practical realities of digital media, where access often requires agreeing to complex contracts that may restrict fair use rights. Moreover, the Digital Millennium Copyright Act (DMCA) allows circumvention of digital locks for lawful fair use purposes.
Courts in recent AI copyright cases, like Kadrey v. Meta and Bartz v. Anthropic, have rejected the idea that unlawful access of source material automatically makes training unfair. This undermines the report’s position and aligns with established legal principles that fair use can apply even when some source materials were obtained without authorization.
Misapplication of the Warhol Decision to AI
A key difficulty in fair use is defining the “use” to be evaluated. The Supreme Court’s Warhol v. Goldsmith ruling highlighted this challenge by examining whether later commercial licensing of unauthorized prints affected their transformative nature.
The Court concluded that commercial use by Warhol’s estate decades after the prints were made could retroactively affect whether the prints were transformative. This creates a paradox where fair use status depends on future commercial decisions, which complicates fair use analysis.
Applying this reasoning to AI is problematic. AI is a general-purpose technology used by many actors for widely varying purposes beyond the original developer’s control. For example, Meta’s Llama model has been adapted for medical research and art conservation—uses far removed from any commercial intent of the training data.
The Copyright Office’s report assumes a single decision-maker controlling training and use, which rarely reflects AI’s actual distributed ecosystem. This leads to confusion about whether a model’s training is fair use if downstream users apply it in ways that compete with copyrighted works. The result is a contradictory view where training can be both fair and unfair depending on unknown future uses.
Practical Impacts of the Report
- The report has had minimal influence on legal proceedings, partly because it was released in a “pre-publication” form amid administrative turnover.
- Recent court decisions have largely ignored the report, as it holds persuasive but not binding authority.
- The American Law Institute’s Restatement of Copyrights, released shortly after, offers a more balanced and accessible treatment of fair use that courts are more likely to consult.
Conclusion
The Copyright Office’s report aimed to provide clarity on AI and fair use but stumbled by endorsing unfounded legal theories, inventing rights without statutory basis, and misapplying Supreme Court precedent. Its attempt to offer categorical answers to nuanced, fact-driven questions falls short. Fortunately, courts appear cautious about relying on the report’s flawed reasoning.
For legal professionals navigating AI and copyright, the best guidance remains grounded in established case law and the evolving judicial interpretation of fair use, rather than the unsettled analyses in this report.
Your membership also unlocks:
