October 13, 2026

Judge Denies Fair Use in First AI Copyright Appeal

By Emily Poler

One thing I find sooooo satisfying is a copyright case where the judge starts by saying, “hey, yeah, the litigants are asking us to figure out infringement related to new technology, but no matter the fancy-sounding jargon you’re throwing at me and what you claim are the earth-shattering stakes, this is JUST a copyright case. And I’m going to apply copyright law the same as I would if you brought a copyright case involving an old-fashioned book.”

Ok, it’s not a direct quote, but it’s the gist. So why does this get me so excited? Regardless of what I think of the outcome of a case, and as much as I like diving into new tech, for me, what I really love about being a lawyer and a litigator is reasoning by analogy. And, in the Third Circuit’s recent decision in Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence, Inc., (which I first wrote about here) the Third Circuit gave me what I love. “Under ROSS’s framing, this case appears to concern the future of AI legal technology. But appearances can be deceiving. In truth, this is no more than an ordinary copyright case,” said Judge Tamika Montgomery-Reeves.

For those of you who aren’t lawyers, we use a Thomson Reuters product called Westlaw for legal research. One of the things that, IMHO, makes Westlaw so useful is its West Key Number System, which is more or less like the Dewey Decimal System (without, at least as far as I know, the racism and misogyny of Melvil Dewey). The system organizes United States law into a structure of topics and key numbers that lawyers use to find cases relevant to specific legal issues and look for other cases that address the same issues. Most cases on Westlaw include a short summary called a headnote, which are generated by Westlaw editors. 

ROSS Intelligence was (it no longer exists) a Westlaw competitor that “created an AI legal search engine that would respond to plain-language legal questions with relevant passages of text from judicial opinions.” To build this, ROSS hired a company to create training data for its AI, which consisted of questions and answers called “Bulk Memos.” Those Bulk Memos were based on — and quoted directly from — Westlaw’s headnotes. ROSS clearly intended its product to be a substitute for Westlaw; it even ran ads comparing its product to the Thomson Reuters database. 

Not surprisingly, Thomson Reuters was none too happy about all this and sued ROSS. In response, ROSS claimed that Westlaw’s headnotes were not subject to copyright protection. It also argued its use of headnotes was an “intermediate” step and referenced a line of cases finding copying of computer code at an intermediate step to be fair use, where such copying was necessary to access purely functional elements of a computer program. 

After some back and forth, the District Court found that headnotes were sufficiently original to be copyrighted. It also denied ROSS’s claim of fair use, noting that ROSS was not copying computer code. Late last month, the Third Circuit affirmed this holding. 

It’s something of a landmark decision, since by my count it is the first case involving an AI platform to be decided by a Federal Appellate Court. And there are a couple of things especially noteworthy about the Court’s conclusions. 

First, since ROSS’s end use — to assist with legal research — was the same as Westlaw’s, the Third Circuit found that ROSS’s use wasn’t particularly transformative, which weighed against a finding of fair use. The Court was also not particularly impressed with ROSS’s reliance on precedent that “intermediate” copying of computer code was fair use because, in contrast to those previous cases, ROSS didn’t NEED to copy the materials. ROSS simply chose to do it because it was the easiest way to train its AI. In addition, ROSS copied headnotes and not computer code.

The Third Circuit also found that, even though Thomson Reuters hadn’t yet licensed its headnotes, ROSS’s conduct could harm the potential market for Thomson Reuters to do so. (When considering fair use courts look at the impact on the potential market for the work being infringed.) This is interesting because earlier this year, the judge in the Meta copyright case found that this factor weighed in favor of fair use because the plaintiffs (authors, in that case) hadn’t presented evidence of harm to the licensing market. Obviously, the Third Circuit saw things differently.

What does Thomson Reuters’ victory mean for other AI-related copyright cases going forward? In my view, its impact may be limited because the use at issue here is so specific, and ROSS’s product was clearly directly competitive with Westlaw. In contrast, the LLMs at issue in cases involving Meta and Anthropic can be used for a wider range of tasks. For example, if an author’s novel is ingested into an LLM that is used to create materials that are not novels — say, an email to opposing counsel — the new use is less clearly competitive. Nonetheless, the ROSS case represents another step in the ongoing journey of AI through the courts, the epic saga of our legal (and cultural) times that we will continue to watch with bated breath.