The Amicus Brief and Its Counterarguments: Why META's Behavior Does Not Constitute Fair Use

by Claude

Introduction

The amicus brief filed by intellectual property law professors Matthew Sag, Zahr K. Said, Jessica Silbey and Rebecca Tushnet in Kadrey v. META Platforms represents a significant effort to defend the use of copyrighted works as training data for generative AI systems. The brief argues that internal copying for the purpose of creating AI tools constitutes fair use, drawing parallels to precedents involving intermediate copying and non-expressive uses. However, a careful examination of recent legal developments, court decisions, and expert analyses reveals substantial flaws in this reasoning that undermine META's fair use defense.

What the Amicus Brief Attempts to Achieve

The amicus brief makes several key arguments to support META's position:

1. Transformative Purpose: The authors argue that internal copying to create AI tools is transformative, similar to cases involving search engines, plagiarism detection, and reverse engineering.

2. Intermediate Copying: They contend that copies made as intermediate steps toward creating a non-infringing output should be considered fair use.

3. Market Impact Limitations: The brief claims that non-expressive, internal uses do not harm cognizable markets for copyrighted works.

4. Technical Non-Expressive Use: The authors suggest that because META's internal copies are never perceived by humans but merely processed by machines, this constitutes non-expressive use.

5. Historical Precedent: The brief cautions against restricting fair use for new technologies, suggesting that past fears about technologies harming copyright owners have proven unfounded.

The amicus brief ultimately seeks to establish that large-scale copying of copyrighted works for AI training falls within fair use protection, regardless of licensing options or commercial applications.

Strong Counterarguments Against the Amicus Brief

1. Warhol v. Goldsmith Has Narrowed Transformative Use

The amicus brief relies heavily on an expansive interpretation of "transformative use" that has been significantly curtailed by the Supreme Court's 2023 decision in Andy Warhol Foundation v. Goldsmith. As Rafael Brown correctly notes in his analysis, this landmark case explicitly ruled that commercial usage that serves "substantially the same purpose" undermines fair use claims. The Court rejected the notion that minor alterations or changes in medium automatically qualify as transformative.

Applying this precedent to META's use of copyrighted works reveals a fatal flaw in the amicus brief's argument. META's AI models are fundamentally commercial products that utilize copyrighted works to generate content that often serves the same purpose as the original works. This directly contradicts the Supreme Court's clarification that transformative use must serve a fundamentally different purpose than the original.

2. Commercial Nature Weighs Strongly Against Fair Use

The amicus brief downplays the commercial nature of META's AI systems, but as Suchir Balaji's analysis demonstrates, this factor significantly undermines META's fair use defense. META's AI models are not academic or research endeavors but commercial products designed to generate billions in revenue.

This commercial nature weighs heavily against fair use, particularly in light of internal documents revealing META's strategic decisions to prioritize commercial gain over proper licensing. The 2025 unsealed emails show META's legal team was directly involved in discussions to stop licensing efforts in favor of using pirated sources, demonstrating willful commercial exploitation rather than good-faith transformative use.

3. Empirical Evidence of Market Harm

The amicus brief dismisses market harm concerns, but empirical research directly contradicts this position. The study "Cloze Encounters: The Impact of Pirated Data Access on LLM Performance" provides concrete evidence that AI models perform measurably better when trained on copyrighted works, with performance improvements of up to 23% when using pirated books. This empirical evidence establishes a direct link between unauthorized use of copyrighted material and commercial benefit.

Additionally, market data shows significant harm to content creators. Stack Overflow experienced a 12% traffic decline after ChatGPT's release, and Chegg's stock dropped 40% due to AI competition. These concrete examples of market substitution directly undermine the amicus brief's claim that AI training has no cognizable market impact.

4. Thomson Reuters v. Ross Intelligence Sets New Precedent

The February 2025 ruling in Thomson Reuters v. Ross Intelligence provides a recent and relevant precedent that contradicts the amicus brief's arguments. The court explicitly rejected fair use for AI training when the resulting system competed with the original market, granting summary judgment against Ross Intelligence for copyright infringement.

While the amicus brief attempts to distinguish this case by claiming it only applies to non-generative AI, the court's reasoning about market harm and commercial purpose applies equally to META's use. The court emphasized that fair use does not protect AI companies when they repurpose copyrighted content in ways that compete with original markets—precisely what META is doing.

5. Systematic Wholesale Copying Exceeds Fair Use Boundaries

The amicus brief attempts to normalize the scale of META's copying, but the industrial-scale, systematic nature of this copying exceeds traditional fair use boundaries. META has copied entire works without permission, not selected portions. Documents reveal that META torrented at least 81.7 terabytes of data, including 35.7 TB from pirated book repositories like Z-Library and LibGen.

This wholesale copying goes far beyond what courts have previously accepted as fair use for intermediate copying. Unlike previous cases involving limited sampling or analysis, META's comprehensive copying represents a fundamental appropriation of copyrighted expression, not just ideas or facts.

6. META's Conduct Demonstrates Bad Faith

While the amicus brief argues that fair use is "not a privilege reserved for the well-behaved," META's documented conduct goes beyond mere unauthorized access. The unsealed emails reveal META employees discussing the risks of being caught and proactively suggesting using VPNs to hide their activities.

Courts have consistently held that deliberate circumvention of access controls and concealment efforts undermine fair use claims. META's documented behavior shows not simply unauthorized access but deliberate efforts to avoid detection and licensing obligations.

7. Alternative Licensing Mechanisms Do Exist

The amicus brief claims that licensing for large-scale AI training is impractical, but META's own actions contradict this assertion. META specifically halted ongoing licensing negotiations in favor of pirated sources, demonstrating that licensing mechanisms do exist but were deliberately avoided.

Moreover, other AI companies have successfully negotiated data licensing agreements with content providers, showing the feasibility of legal approaches. The amicus brief's arguments about transaction costs and impracticality are belied by the existence of these successful arrangements.

8. Limited Applicability of Search Engine Precedents

The amicus brief heavily relies on search engine cases like Authors Guild v. Googleand Perfect 10 v. Amazon, but these cases are distinguishable in critical ways. Search engines direct users to original content, while META's AI can generate substitutes for the original works.

Furthermore, META's AI extracts not just facts or ideas (which are not protected by copyright) but creative expression itself—the very substance copyright law is designed to protect. As the 2024 AI copyright analysis shows, AI systems trained on copyrighted works can reproduce substantial portions of those works, undermining the non-expressive use argument.

Conclusion

The amicus brief filed by IP law professors attempts to extend fair use doctrine to protect META's wholesale copying of copyrighted works for commercial AI training. However, recent legal precedents, empirical evidence of market harm, and META's documented conduct provide compelling reasons to reject this interpretation.

META's behavior does not constitute fair use because:

• It fails the transformative use test established in Warhol v. Goldsmith

• It is fundamentally commercial in nature

• It demonstrably harms existing markets for copyrighted works

• It represents systematic, wholesale copying rather than limited sampling

• It involves deliberate efforts to conceal copyright violations

• It rejected available licensing options in favor of unauthorized use

The proper application of fair use doctrine requires balancing the interests of copyright holders with the advancement of knowledge and creativity. META's behavior disrupts this balance by appropriating creators' works without permission or compensation while generating significant commercial benefit. This is precisely the type of market interference that copyright law is designed to prevent.