On Tuesday, February 11, a Delaware district court issued much-awaited summary-judgment decisions in Thomson Reuters Enterprise Centre GmbH et al v. ROSS Intelligence Inc., No. 1:20-cv-613, potentially shaping how future courts will analyze transformative use and intermediate copying of copyrighted works used to train artificial intelligence models. The decisions generally favor the copyright holders but appear to spare generative AI defendants.
Background
Thomson Reuters owns registered copyrights in Westlaw’s legal materials, including headnotes and its Key Number System that summarize key points of law and case holdings. Ross Intelligence developed AI-based legal research software, using 25,000 “Bulk Memos” from LegalEase as training data. These Memos were derived from Westlaw headnotes.
In 2023, Thomson Reuters requested summary judgment on copyright infringement, while both parties filed cross motions on the fair-use defense. Initially denied, the motions were later reconsidered by the court, which resulted in this week’s decision.
Direct Copyright Infringement
The court granted partial summary judgment to Thomson Reuters on direct copyright infringement. Noting the extremely low threshold for originality, the court upheld the validity of Thomson Reuters’s copyright in the headnotes and Key Number System, analogizing the crafting of a headnote from a judicial opinion to sculpting a block of raw marble. Nevertheless, determining which individual headnotes fall within the copyright term requires a jury trial.
The court also found evidence of actual copying for a subset of headnotes based on expert opinion and textual similarities. The rest of the headnotes must go to trial.
Fair-Use Defense
Balancing the four fair-use factors and giving most weight to factor four, the court ruled that Ross’s use of the copyrighted works is not fair, even though factors two and three favored Ross.
Applying the Supreme Court’s Andy Warhol analysis to asses factor one, the court found Ross’s use was not transformative because it aimed to develop a competing legal research tool. The court distinguished Ross’s copying from permissible intermediate copying in software cases, ruling that copying computer code was not at issue and copying was not necessary for innovation.
Factor four favored Thomson Reuters as Ross “meant to compete with Westlaw by developing a market substitute.” The court dismissed claims of public benefit, as legal opinions are freely available, but Westlaw’s curated content is not.
Key Takeaways
Parties in AI-related copyright disputes should take note of the court’s fair-use analysis, which appears to collapse the first and fourth fair-use factors. Notably, transformative use—the go-to defense for AI defendants—may not hold much water when the defendant’s product competes with the copyright holder. And intermediate copying may be relegated to specific factual scenarios.
Further, the court raises the prospect that the fourth fair-use factor could be met by a potential market for AI training data. This increases an AI defendant’s potential to fail the fourth factor when copyrighted material is part of its training data.
Although the court was careful to exclude generative AI from its analysis by distinguishing Ross’s AI tool, which simply outputs judicial opinion in response to a user’s legal question, from generative AI capable of writing new content, the court’s application of the Warhol framework could nevertheless apply to generative AI.
