Meta Wins Kadrey Case, but Judge Opens New AI Copyright Front

On June 25, 2025, Judge Vince Chhabria granted Meta summary judgment on the theory presented by those plaintiffs. The original source supports the documentary core of the event; the ruling expressly limits its reach and does not declare all training on copyrighted works lawful.

The decision, issued on June 25, does not make training language models on copyrighted works automatically legal. On the contrary, the judge delivered a clear warning to AI companies. The authors largely lost because they did not provide enough evidence of how the models could damage the book market.

Llama’s training was deemed transformative

The lawsuit, Kadrey v. Meta, was brought by 13 authors, including novelist Richard Kadrey. They accused Meta of copying their books without permission to train Llama, its family of open-source language models. The texts came from so-called shadow libraries: online repositories that distribute copyrighted works for free without authorization from their owners. Document supporting the figure.

Meta did not deny using copyrighted books. The question was whether that use fit within the four factors U.S. law applies to determine fair use: the purpose of the use, the nature of the work, the amount copied, and the effect on its market.

Chhabria sided with Meta on three of the four factors. He found that training a model has a “highly transformative” purpose: the books were created to be read, while the company used them to develop a system capable of processing and generating language.

The judge did acknowledge that novels, memoirs, and plays are “highly expressive” creations, a circumstance that favors the authors. But he found it reasonable for Meta to copy the works in full for its training objective. He also rejected the plaintiffs’ claim that Llama had been shown to reproduce their books in a way that would substitute for the originals.

The relevant market is not just the licensing market

The most important part of the ruling concerns the fourth factor: market harm. Among other arguments, the writers claimed that Meta had deprived them of a potential market for licensing their works to train AI. The court rejected that theory. The mere existence of a hypothetical market for licensing data is not enough, by itself, to bar fair use.

But Chhabria identified a different, potentially stronger path: market dilution. A model trained on millions of books could enable the rapid production of countless works that compete with the originals for readers, attention, or revenue, even if those new works do not literally infringe the copyright in any specific text.

The judge warned: “No matter how transformative LLM training may be, it’s hard to imagine that it can be fair use to use copyrighted books to develop a tool to make billions or trillions of dollars while enabling the creation of a potentially endless stream of competing works that could significantly harm the market for those books.”

That theory did not save the lawsuit, however. The authors presented a market-dilution theory that was too weak and lacked enough concrete evidence to take the case to trial. That is why the ruling is limited. The judge described the decision as limited: it does not declare the use of copyrighted works to train models generally lawful. Instead, it concludes that these plaintiffs advanced the wrong arguments and failed to develop evidence for the appropriate theory.

A different victory from Anthropic’s

The ruling comes two days after Judge William Alsup’s decision in the case against Anthropic. Alsup also deemed language-model training transformative, but legally separated several actions: training, digitizing printed books, and creating a central library of pirated copies.

Chhabria took a different approach. He analyzed Meta’s downloads according to their ultimate purpose—the training of Llama—rather than treating them as independent uses. That divergence matters because AI companies face not only the question of what a model does with a work, but also how they obtained the copy and what infrastructure they built with it.

For publishers, writers, and AI companies, the Kadrey case shifts the next battleground. The debate is no longer limited to whether a model memorizes or regurgitates protected passages. Future lawsuits will likely try to determine whether generative systems create a substitute supply broad enough to reduce the commercial value of human-made works.

That showing will be difficult: it will require data on sales, user behavior, and actual competition—not merely the abstract possibility that an AI could write a similar book. But the ruling makes clear that if such data reaches a courtroom, the transformative nature of training may not be enough to protect an AI company.

Turning the headline into a check

The document heading says what exists: complaint, motion, order, proposal or judgment. Each word changes what a reader may do. An allegation is unproven; a proposed settlement is unapproved; an order may resolve only one motion. the ruling expressly limits its reach and does not declare all training on copyrighted works lawful. Before extracting a rule, locate the operative text, date and affected parties.

Break the case down by conduct and claim. Obtaining a copy, retaining it, training with it and generating an output are different acts; a court may treat them differently. The evidence and procedural stage matter too. Summarizing everything as 'won' or 'lost' erases the very boundary that makes the ruling usable.

To assess how to read party, claim, evidentiary record and scope before turning a ruling into a general rule, build a timeline: filing, response, ruling, possible appeal and pending conditions. Then ask which remedy was requested and which was granted. Maximum damages, a proposed fund and money received are not equivalent either. Dates prevent an early document from being credited with what happened later.

What the record must preserve

A United States ruling does not automatically become a universal rule. Jurisdiction, governing law, facts and parties limit its reach. An article can teach document reading but cannot replace advice on a particular case. The useful skill is knowing which question to take to a professional and which page supports each premise.

An evidence sheet separates four columns: what the source claims, what it shows, what it did not measure and what would change the conclusion. That discipline prevents an absence from becoming a promise and a condition from vanishing in summary. It also lets the story be updated without rewriting history from a later outcome.

Include a negative case before deciding. Find a situation where the system, rule, transaction or study does not meet the need and record the signal that would require stopping. Selected successes show that something can happen; the negative case reveals the boundary and lowers the cost of discovering it after deployment.

The skill that outlasts the announcement

A valid comparison preserves denominator and axis. It does not pit a point figure against an average, future capacity against installed capacity or a forecast against an observation. When two sources use similar language, reconstruct what they counted and over what period. If those differ, publish them as different measures instead of inventing a ranking.

The record should survive a version change. Keep URL, consultation date, document, configuration and decision. When new evidence appears, add it with its date and explain what it changes. That traceability prevents opposite errors: keeping an expired conclusion or pretending later information was known on the event date.

The transferable skill in this story is how to read party, claim, evidentiary record and scope before turning a ruling into a general rule. The procedure is short: name the document, preserve the date, fix the axis, find the condition and design a check that can fail. With those steps, a reader need not accept or reject the announcement by intuition; the decision follows a visible chain of evidence.