Introduction
In June 2025 a US judge ruled for Meta in a copyright suit brought by authors, and many reports summarized the result as a court approving AI training. The judge had written, in the order itself, that it meant no such thing.
This reading goes through the passage where he says what his ruling does and doesn't decide, to show how a court can find for one side and warn it in the same document.
Locating the Passage
The document is the order of June 25, 2025, in Kadrey v. Meta Platforms, Inc., by Judge Vince Chhabria of the US District Court for the Northern District of California. Thirteen authors had sued Meta for downloading their books from online "shadow libraries," sites that host pirated copies, and using them to train its language models. The order is free on FindLaw, and the link is in the References.
The passage is the last three paragraphs of the order's opening section, which comes before the detailed background and analysis. This reading also draws on the order's concluding section. The order has no numbered paragraphs, so the reading identifies each quotation by where it sits. All quotations are from the order (Kadrey v. Meta 2025).
Fair use is the US doctrine that permits some uses of a copyrighted work without the owner's permission. Courts weigh four factors, the fourth being the effect of the use on the market for the work. This reading describes a court document. It isn't legal advice.
Walking Through the Passage
Step 1: Read what the judge says the ruling doesn't stand for
The final paragraph of the opening section says: "this ruling does not stand for the proposition that Meta's use of copyrighted materials to train its language models is lawful."
"Stand for the proposition" is lawyers' language for the rule that a decision establishes. The judge is telling future readers which rule his decision does not establish. It doesn't establish that Meta's training was lawful.
The sentence before it narrows the ruling further: "This is not a class action, so the ruling only affects the rights of these thirteen authors." Everyone else whose work Meta used keeps whatever claims they have.
Step 2: Read what he says it does stand for
The next sentence is: "It stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one."
A "record" is the body of evidence the parties put before the court. The sentence makes two criticisms of the authors' case. They argued the wrong things, and they brought no evidence for the argument that might have worked.
The paragraph before names the wrong arguments. The authors said the model could reproduce small snippets of their books. They also said that unlicensed training cost them the chance to license their books for training. The judge writes that "both of these arguments are clear losers." On the first, the model couldn't produce enough of their text to matter. On the second, he held that authors aren't entitled to the market for licensing their works as training data.
Step 3: Find his description of market dilution
The same paragraph names the right argument. He calls it "the potentially winning argument": that Meta copied the books to create "a product that will likely flood the market with similar works, causing market dilution."
Market dilution means that a market for human-made works is weakened when large volumes of AI-generated works of the same kind compete with them. Earlier in the opening section the judge explains why he takes it seriously. Generative AI, he writes, "has the potential to flood the market with endless amounts of images, songs, articles, books, and more," made with "a tiny fraction of the time and creativity that would otherwise be required."
On this argument the authors offered almost nothing. In his words, they "barely give this issue lip service, and they present no evidence about how the current or expected outputs from Meta's models would dilute the market for their own works."
Step 4: Ask why a judge would rule for one side and warn it in the same order
The answer is in the paragraph that begins the turn to this case: "Courts can't decide cases based on general understandings. They must decide cases based on the evidence presented by the parties."
The judge holds a general view, and he states it. Just before that turn he writes that "in many circumstances it will be illegal to copy copyright-protected works to train generative AI models without permission." He also holds that a court may act only on what the parties prove. Meta brought evidence on market effects and the authors didn't, so on this record Meta had to win.
The warning serves two audiences. It tells other AI developers not to treat the result as clearance. And it tells future plaintiffs which argument to build. A judge who thought the law favored developers in general would have had no reason to write it.
Step 5: Say what a future plaintiff would need to show
The order's concluding section spells this out. Because Meta's use was highly transformative, the authors "needed to win decisively on the fourth factor." The judge adds that if they had "presented any evidence that a jury could use to find in their favor on the issue, factor four would have needed to go to a jury."
So a future plaintiff would need evidence that the outputs of the defendant's model compete with the plaintiff's own kind of work and reduce its market. The order suggests where that is more and less likely. News articles and typical genre fiction may be vulnerable to AI-generated substitutes. A memoir, which people read because of who wrote it, may not be.
He also predicts how such cases will go: "it seems like the plaintiffs will often win, at least where those cases have better-developed records on the market effects of the defendant's use." That is a forecast by one trial judge, and it binds nobody.
Key Considerations
The order decides motions for summary judgment. That is a ruling made without a trial, on the ground that the evidence each side has presented leaves nothing for a jury to decide. It tests the evidence in the file. A different file could produce a different result under the same law.
The common mistake is to report the case as "court says AI training is legal." The order says the opposite about its own meaning. What it decides is that thirteen authors didn't prove their case against one company.
The reverse mistake is to report it as a finding that training is illegal. The judge's general statements about what "will likely" be unlawful are his reasoning, and they weren't needed to decide the case. They don't bind other courts. Two days earlier another judge of the same court, in a case against Anthropic, had treated the market-dilution concern as outside what copyright protects, and Judge Chhabria's order criticizes that reasoning by name.
The order also left part of the case undecided. A separate claim, that Meta distributed the authors' books in the course of downloading them, wasn't covered by the motions and remained live. A trial-court ruling of this kind can also be appealed.
Summary
The order grants judgment to Meta and, in the same passage, limits what that judgment means. Stated plainly:
The holding. On the evidence these thirteen authors presented, Meta was entitled to judgment that its use of their books to train its language models was a fair use. The authors relied on two arguments the judge rejected and offered no meaningful evidence on market dilution, the argument he considered potentially winning.
The warning. The ruling does not establish that such training is lawful, and a plaintiff who proves that a model's outputs dilute the market for their work could win.
References
- Kadrey v. Meta Platforms, Inc. 2025. No. 23-cv-03417-VC, US District Court, Northern District of California, June 25, 2025.