Background
Thomson Reuters operates Westlaw, whose editors prepare headnotes that identify significant legal propositions in judicial opinions and summarize them in concise, standalone language. Although judicial opinions themselves are not copyrighted, Thomson Reuters holds registered copyrights covering Westlaw content, including its headnotes.
ROSS Intelligence developed a commercial AI-powered legal-research platform designed to answer plain-language questions with relevant passages from judicial opinions. Its contractor created approximately 25,000 training memoranda using Westlaw headnotes to formulate legal questions and linked opinion passages as answers. Thomson Reuters sued for copyright infringement, and the District Court granted partial summary judgment after determining that 2,243 headnotes had been copied and that ROSS’s use was not fair. The court certified the originality and fair-use issues for interlocutory appeal.
The Court’s Holding
The Third Circuit affirmed. It held that the 2,243 headnotes were sufficiently original because Thomson Reuters’s editors exercised creative judgment in selecting which legal points to feature and deciding how to express them accurately, concisely, and with enough context to stand alone. Copyright protection for those headnotes did not monopolize the law because judicial opinions remained freely available, and the merger doctrine did not apply because points of law can be expressed in many ways.
The court also held that ROSS failed to establish fair use. Its use was commercial and, at most, minimally transformative because both companies used the headnotes toward the same ultimate purpose: providing a legal-research platform that helps users locate responsive legal authority. ROSS copied entire headnotes even though it could have trained its system using freely available judicial opinions, and its competing product diminished the headnotes’ value to Westlaw while usurping a developing market for licensing headnotes as AI-training data. Although the factual nature and prior publication of the headnotes caused the second fair-use factor to favor ROSS slightly, the other three factors weighed against fair use.
Key Takeaways
- Editorial headnotes can satisfy copyright’s low originality threshold when editors make creative choices about which legal propositions to include and how to express them.
- Using copyrighted material as an intermediate step in AI training is not automatically transformative; the court examined the ultimate purpose of the copying and found that ROSS sought to build a commercial substitute for Westlaw.
- Market harm may include reduced value of copyrighted content as a feature of a larger product and displacement of a developing licensing market for AI-training data.
Why It Matters
The decision applies traditional copyright principles to AI training and makes clear that labeling copying as part of an AI-development process does not itself establish fair use. Where copyrighted material is used to build a directly competing product serving substantially the same purpose, commerciality, extensive copying, and market substitution may weigh heavily against the defense.
The court emphasized, however, that its ruling concerned ROSS’s nongenerative legal-search system and the specific record before it. It distinguished generative-AI cases involving systems capable of producing original expression and did not announce a categorical rule governing all AI training.