A federal court ruling is sure to spawn a thousand analogies in future lawsuits.
When is training artificial intelligence on somebody else’s work a copyright violation? The Third Circuit federal court of appeals decided to draw a line in Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence, Inc. Whether or not this is the last stop for this suit, the case may well prove influential precisely because lawyers and judges are intimately familiar with the copyrighted text in question.
Judicial opinions can’t be copyrighted, because they are official government documents. But from the earliest days of the republic, they were published not by the courts but in privately compiled law books sold for profit. West Publishing, for many years, has published volumes of opinions with numbers that become their official citations, and has used a proprietary “keynote” system (complete with its own Dewey Decimal System–like numerical classifications by topic) for summarizing at the front of an opinion its conclusions on various points of law. In its books and later its Westlaw system for computerized legal research (which itself has been around since the early 1990s), West has always taken very seriously the unique, copyrighted value that these headnotes (what West brands as “keynotes”) add to the opinions themselves. Indeed, the Supreme Court noted as far back as Callaghan v. Myers (1888) that judicial opinions are not copyrightable but that a private publisher can get “a copyright for the volume which will cover the matter which is the result of his intellectual labor” such as “head-notes.”
The keynotes are, as the court observes, the specialized work of West:
In Westlaw’s user interface, the headnotes sit above the opinion, and clicking on a headnote takes the Westlaw user to the corresponding passage of the opinion upon which the headnote is based. . . . When drafting headnotes, Thomson Reuters’s editors must follow precise guidelines. They must make each headnote clear and concise, including only the essential facts for the legal contention or holding at issue. They are directed to limit each headnote to 800 characters, where possible.
A competitor, ROSS Intelligence, Inc., tried not only to use an AI to do the same job, but trained it on West’s own keynotes:
ROSS . . . created an AI legal search engine that would respond to plain-language legal questions with relevant passages of text from judicial opinions. ROSS had access to a bank of approximately ten million uncopyrighted judicial opinions from which it could respond with an excerpt relevant to a user’s query. ROSS’s AI was not a generative AI, meaning it would not create any new expression; it would only return text passages from preexisting judicial opinions.
For ROSS’s AI legal search engine to work, the AI program had to learn what made a judicial opinion responsive to a user’s legal question. To train its AI program to do so, ROSS enlisted a company called LegalEase Solutions (“LegalEase”) to create AI training memoranda. LegalEase wrote approximately 25,000 legal memos, each of which presented a legal question and four to six judicial opinion passages in response. Memo-drafters for LegalEase, and its subcontractor Morae Global, used thousands of Westlaw headnotes to create the memos’ legal questions, as the headnotes provided “an easy way [to] fram[e] questions.” . . . The memos’ responses to the legal questions needed to include at least one great, good, topical, and irrelevant judicial opinion passage, labeled as such. A great passage was completely responsive; a good passage was mostly responsive; a topical passage was somewhat responsive; and an irrelevant passage was not responsive. The “[g]reat” passages were “[m]ost frequently” the judicial opinion passages that Westlaw linked to the headnote that ROSS used as the memo’s legal question. . . . These memos were then converted into a machine-readable format and used to train the ROSS AI program to identify patterns regarding which judicial opinion passages respond well to legal questions. [Emphasis added; citations omitted.]
This, the court concluded, was not fair use of copyrighted material:
ROSS counters that Thomson Reuters seeks to squelch innovative AI technology. But new technology is often at the forefront of our copyright cases. . . . 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; we must determine whether Thomson Reuters’s materials are sufficiently original for copyright protection and whether ROSS’s use of those materials was fair. The undisputed evidence reflects that Thomson Reuters’s materials possess a creative spark, and ROSS aspired to be a direct competitor by using them for a highly similar purpose. We therefore hold that Thomson Reuters’s materials are copyrightable and that ROSS’s use was not fair. . . .
True, ROSS did not use the headnotes in precisely the same fashion as Thomson Reuters. ROSS took an intermediate step of using the headnotes to train an AI program, which arguably presents a slight degree of difference in use. But the undisputed evidence demonstrates that ROSS used the headnotes to train an AI program for the benefit of its legal-research platform.
The court also rejected an analogy to cases where computer code was copied just to allow software to interface with an existing program:
ROSS does not need to copy Thomson Reuters’s headnotes to access the underlying unprotected information. ROSS had access to the underlying judicial opinions and could freely copy them to make the memos needed to train its AI. It chose not to do so because copying the headnotes offered an “easy” way to create its training memos.
Many authors and publishers won’t be able to present this sort of slam-dunk case that a direct competitor is using AI to steal their own skills at summarization of the same information. But the framework that the court used here to find that AI training had gone too far is doubtless one that will spawn a thousand analogies in future lawsuits. And maybe even its own keynote.