While Meta secured a win, the court's opinion underscores that the victory was procedural and evidentiary—not a definitive declaration that AI training on copyrighted works is inherently lawful.

The Legal and Strategic Implications of the Kadrey v. Meta Fair Use Ruling

by ChatGPT-40

A federal court decision in Kadrey v. Meta Platforms, Inc. has dealt a significant blow to authors and copyright owners seeking to assert control over the use of their works for AI training. While Meta secured a win, the court's opinion, written by Judge Vince Chhabria, underscores that the victory was procedural and evidentiary—not a definitive declaration that AI training on copyrighted works is inherently lawful. The ruling exposes strategic weaknesses in the plaintiffs’ legal approach and offers critical lessons for other copyright holders engaged in or considering similar lawsuits.

1. What the Court Actually Ruled

Contrary to headlines suggesting a broad endorsement of AI training on copyrighted works, the court’s opinion is nuanced and cautionary. The ruling states explicitly:

Judge Chhabria dismissed the plaintiffs’ claims not because AI training was found universally lawful under fair use, but because the authors failed to provide sufficient evidence that Meta’s use of their books caused market harm—a key factor in fair use analysis.

2. Why the Plaintiffs Lost

The plaintiffs (including well-known authors like Richard Kadrey and Ta-Nehisi Coates) advanced two main arguments:

• Snippets and Memorization: That Meta’s LLaMA models could regurgitate snippets from their books.

• Licensing Market Harm: That Meta diminished a potential market for licensing their works for AI training.

Judge Chhabria found both arguments unpersuasive:

• Meta’s models could only output small portions (e.g., 50 tokens or fewer) of the plaintiffs’ works, even under adversarial prompting—not enough to constitute a meaningful market substitute.

• The plaintiffs failed to establish that a viable market for licensing books for AI training existed or was harmed in any concrete way.

Crucially, the plaintiffs “barely gave lip service” to the potentially strongest argument: that generative AI, by being trained on copyrighted books, could flood the market with similar works, undermining demand for original human-created content. They also failed to present data on how Meta’s models actually produce outputs that would displace or dilute the market for the plaintiffs’ works.

3. Consequences for Other Court Cases

This ruling is highly instructive but not binding on other courts or plaintiffs. Its precedential value is limited to the Northern District of California, and even there, it does not foreclose similar claims brought with better evidence and strategy.

However, the decision does signal a high burden of proof for plaintiffs in AI copyright cases. Courts appear open to recognizing that AI training could cause economic harm, but plaintiffs must:

• Quantify and demonstrate actual or likely market harm.

• Identify a relevant and developed licensing market for training data.

• Produce outputs showing clear substitution effects.

Without this evidentiary foundation, future lawsuits are likely to meet the same fate.

Furthermore, the ruling directly contrasts with Judge Alsup’s recent opinion in Bartz v. Anthropic, which emphasized the transformative nature of AI training and minimized concerns over market harm. Judge Chhabria criticized this view as an “inapt analogy,” suggesting a judicial split on how to weigh the fair use factors in AI cases—a factor that may prompt appellate review and eventual Supreme Court clarification.

4. What Plaintiffs Should Be Doing Next

a. Refile with Better Evidence (If Possible)

The court granted summary judgment to Meta, but only as to the named plaintiffs’ claims. Others may still bring similar cases. Key improvements for future lawsuits should include:

• Empirical studies or expert reports showing market displacement by LLM outputs.

• Demonstrations of outputs that closely mimic or replicate the protected works.

• Evidence of licensing interest or actual negotiations with AI developers for training rights.

b. Bring Class Actions with Stronger Claimants

A broader class of plaintiffs—potentially including textbook publishers, reference book authors, or other high-value rights holders—could present a more compelling market harm argument, particularly where LLMs are shown to produce substitutive content.

c. Focus on Market Dilution and Substitution

As Judge Chhabria emphasized, market dilution is the most “potentially winning” theory. Plaintiffs must argue and prove that:

• The AI outputs function as substitutes for their works in specific genres or markets.

• This substitution undermines the economic incentive for human authorship.

d. Support Legislative Reform and Licensing Frameworks

Outside the courtroom, plaintiffs should lobby for new legislation clarifying AI’s use of copyrighted materials and create or join collective licensing schemes, which the court noted are currently lacking. Without such mechanisms, courts may continue to rule that no recognizable market harm exists—even if creators feel exploited.

5. Strategic Recommendations for Rights Holders

• Form licensing collectives to aggregate rights and create an identifiable market.

• Partner with technologists to test and document model outputs for infringing use.

• Develop market substitution models showing economic loss or devaluation.

• Focus on infringement by distribution, not just ingestion, including torrent-based acquisition or unauthorized propagation of training data.

• Leverage ongoing DMCA and fraud-based claims, where appropriate.

Conclusion

While Meta’s victory in this case marks a temporary setback for rights holders, the ruling is a clarion call rather than a death knell. It makes clear that fair use in the AI context will turn not on rhetoric but on rigorous evidence of market impact. Future plaintiffs who heed this lesson—bringing stronger claims, supported by concrete data and framed around market substitution—may still succeed where these authors did not. The litigation battleground is shifting, and the next moves will be decisive in shaping how copyright law responds to the age of generative AI.