This session examines the clash between copyright law and artificial intelligence through the lens of Thomson Reuters v. ROSS Intelligence. We will read Judge Bibas’s 2025 decision and ROSS’s Opening Brief on appeal to the Third Circuit. The case asks whether Westlaw headnotes are copyrightable and whether their use to train a legal search AI is protected by fair use. The materials set up a rich contrast: the court frames ROSS as a competitor free-riding on Westlaw’s summaries, while ROSS presents itself as building a transformative AI search tool that only outputs public-domain judicial opinions.
ROSS built a legal search engine that worked in two stages: first, keyword search over a database of judicial opinions, and second, AI re-ranking of results using a machine learning model (LambdaMART) trained on question–answer sets derived from Westlaw headnotes. The product never displayed headnotes; instead it ranked opinion passages by how well they answered a lawyer’s query. The infringement claim arose because the training data — “Bulk Memos” produced by LegalEase from Westlaw headnotes — copied and adapted thousands of headnotes without permission.
Discussion Focus:
- Where should the line be drawn between originality in factual legal summaries and public-domain judicial opinions?
- Is AI training on protected works more like Google Books (transformative indexing) or Warhol (competing repurposing)?
- What do the technical details of ROSS’s system tell us about how courts should evaluate “transformation” in AI search?
- How might these doctrines affect the design of legal tech search tools, balancing innovation with copyright boundaries?
Please see below link to case materials which is assumed reading in order to participate in the discussion:
ROSS-Opening-Brief-FILED-ca3.125346.28.0_1
Judge-Bibas-SJ-order-rejects-fair-use-defense
Discussion led by Adrian Cartland