\n\n
A federal appellate court has delivered the first major ruling on whether copyrighted material can be used to train generative AI. On September 29, 2026, the Third Circuit affirmed a lower‑court finding that Ross Intelligence infringed Thomson Reuters’ Westlaw headnotes by feeding them into its AI legal‑research tool, concluding that the headnotes are protected by copyright and that Ross’s use failed the fair‑use test. The court treated the dispute as a standard copyright case, noting that the headnotes contain enough original expression to merit protection and that three of the four fair‑use factors—purpose of use, amount copied, and market harm—favored the plaintiff.
The decision has immediate repercussions for AI developers, signaling that the ordinary copyright framework applies to training data and that merely training a model does not automatically qualify as fair use. Companies must now assess whether their data sources are copyrighted, seek licenses, or rely on public‑domain material to avoid infringement.
Meanwhile, the publishing industry is quietly expanding its own AI use. Interviews with anonymous staff at three major houses reveal that tools such as Claude and ChatGPT are being deployed to draft publicity copy, back‑cover text, and even generate cover art, often without author disclosure. This internal shift occurs as publishers continue to litigate over AI misuse—most notably a joint lawsuit against Meta for alleged copyright violations and the cancellation of high‑value deals when AI‑generated manuscripts are suspected. The sector also faces external pressure: Anthropic settled a $1.5 billion class action alleging it trained Claude on pirated books, and OpenAI and Microsoft successfully struck a critical report in a multidistrict copyright case.
Why it matters: The Third Circuit ruling clarifies that AI training is not a blanket exemption from copyright law, urging firms to adopt rigorous data‑licensing strategies. At the same time, publishers’ growing reliance on LLMs for marketing and design highlights a broader industry trend of embedding AI deeper into content workflows, raising new questions about transparency, author rights, and potential further litigation.
What to watch next:
- Upcoming fair‑use challenges in cases like Concord Music Group’s suit against Anthropic over song‑lyric training data.
- Potential legislative moves to require AI developers to disclose copyrighted training material.
- How major publishers balance AI‑driven efficiency with author consent and copyright compliance as internal AI use expands. [[1, 2, 3]]