This post was originally published on this site.
“Wisdom does not always find me, so I try to embrace it when it does–even if it comes late, as it did here.” Judge Stephanos Bibas
Yesterday, Circuit Judge Stephanos Bibas, sitting by designation in the District of Delaware, issued a ruling updating a previous summary judgment decision dismissing copyright infringement allegations made by Westlaw legal research service provider Thomson Reuters against a competing artificial intelligence (AI) search tool developed by Ross Intelligence. Among the top reconsiderations in Judge Bibas’ recent decision is his fair use analysis, which now recognizes the non-transformative nature of Ross’ use of copyrighted headnotes that summarize legal decisions.
In September 2023, Judge Bibas issued a summary judgment ruling largely dismissing motions from both parties, concluding that most disputed issues had to be submitted to a jury. Among those issues was the breadth and validity of Thomson Reuters’ alleged copyright in its compilation registration for Westlaw headnotes, and Ross’ fair use defense. Although liability was ultimately left undecided by Judge Bibas on summary judgment, his ruling indicated that Ross’ use could be analogous to those from so-called “intermediate copying” case law, where copying of software source code was deemed transformative.
Westlaw Legal Headnotes Meet Low Threshold for Originality
“Wisdom does not always find me, so I try to embrace it when it does––even if it comes late, as it did here,” wrote Judge Bibas in this week’s updated summary judgment ruling. Studying case materials closely leading up to the August 2024 trial date, Judge Bibas invited new briefing from both parties after realizing that his first decision “had not gone far enough.” The new decision now grants most of Thomson Reuters’ motion for partial summary judgment on copyright infringement and fair use, while also denying Ross’ motion for summary judgment on fair use.
“The law is no longer a brooding omnipresence in the sky; it now dwells in legal research platforms,” wrote Judge Bibas, noting that Westlaw is the largest platform available online for accessing case law, statutes and related summaries. Contracting with legal services company LegalEase, Ross purchased about 25,000 bulk memos containing legal questions with good and bad answers developed from Westlaw headnotes and used those memos to train its AI search tool.
Comparing the bulk memo questions and the Westlaw headnotes side by side, Judge Bibas concluded that Ross infringed a total of 2,243 headnotes, the only remaining issue being whether copyright in those headnotes had expired. Revising his opinion from the 2023 summary judgment ruling, Judge Bibas now found enough originality in Westlaw’s headnotes to find no genuine dispute on that fact.
“More than that, each headnote is an individual, copyrightable work. That became clear to me once I analogized the lawyer’s editorial judgment to that of a sculptor. A block of raw marble, like a judicial opinion, is not copyrightable. Yet a sculptor creates a sculpture by choosing what to cut away and what to leave in place.” – Judge Bibas
Bibas also found originality in Westlaw’s Key Number System for categorizing legal holdings, but noted that factual disputes remained as to Ross’ use of that system. The opinion found actual copying involved with 80% of the Westlaw headnotes at issue on summary judgment, noting that 20% of the batch, or 587 headnotes, were used for later validation and not the infringing AI model training. He also found substantial similarity between Ross’ bulk memo questions, which closely tracked the language of the Westlaw headnotes at issue. Ross’ raised several copyright defenses, including innocent infringement and scenes à faire, that failed as a matter in this revised opinion.
No Transformative Use, Potential Market Impacts Derail Ross’ Fair Use Defense
Moving on to fair use, Judge Bibas found that Thomson Reuters prevailed on the doctrine’s application thanks in large part to reconsidering the application of the U.S. Supreme Court’s holding in Google v. Oracle America (2021) and Ninth Circuit case law on fair use in the context of analyzing computer programs. In those intermediate copying cases, the alleged defendant succeeded on fair use because computer code was copied in order to reverse engineer the uncopyrightable functionality of those software programs. Here, under the Supreme Court’s revised fair use standard from Andy Warhol Foundation for the Arts v. Goldsmith (2023), the broad purpose and character of Ross’ use created a competing legal service that was not transformative.
Factor two, the nature of Thomson Reuter’s original work, weighed in Ross’ favor with the caveat that this factor rarely plays a significant role in fair use determinations. Fair use factor three, the substantiality of the infringed portion of the copyrighted work, also weighed in Ross’ favor because Ross’ use in training its AI search tool did not make Westlaw headnotes available to the public. But factor four, “undoubtedly the single most important element of fair use” because it assesses impacts to the market for the original work, fell in Thomson Reuters’ favor. No longer a transformative use, Ross’ competing legal service product created enough of a potential impact on the market for AI training data to outweigh the public interest in access to the law asserted by Ross.
Judge Bibas’ decision did point out that this fair use analysis was confined to the non-generative AI service at issue in the case, which he chose to note due to rapid changes in the AI landscape. Still, this ruling will likely impact several high-profile copyright infringement cases between creators and generative AI developers currently winding their way through U.S. district court.
Commenting on the case, Randy McCarthy of Hall Estill said in an emailed statement that, while the decision can still be overturned and is just “one battle in a larger war,” Bibas’ analysis of the fourth fair use factor could be significant. McCarthy explained:
“The most important aspect of the opinion, to me, is the discussion of the fourth element of the fair use analysis. The Judge identifies this as the most important factor, and focuses upon the impacts upon the market for the original work. This is going to be heralded by existing groups of artists and content creators as the key to their case against the other generative AI systems.”