The news
On September 29, 2026, the U.S. Court of Appeals for the Third Circuit ruled that Ross Intelligence's use of 2,243 Westlaw headnotes to train its legal search engine was not fair use, the first federal appeals decision on AI training fair use, according to LawNext.
Thomson Reuters (TRI), which owns the Westlaw legal research service, sued Ross for copyright infringement. Headnotes are short editorial notes that summarize the points of law in a court opinion. Circuit Judge Montgomery-Reeves wrote the opinion for a panel with Judges Restrepo and Bove, affirming partial summary judgment granted by Judge Stephanos Bibas, a Third Circuit judge who heard the case in Delaware federal court by designation.
According to the opinion, Ross built a search engine that answered plain-language legal questions with passages from about 10 million judicial opinions; it did not generate new text. To train it, Ross hired LegalEase Solutions, which wrote about 25,000 training memos, and the memo writers used Westlaw headnotes to frame the questions. Ross advertised its product against Westlaw at comparable prices, and some law firms switched.
The panel first held that the headnotes are original enough for copyright, because editors chose which points of law mattered and how to word them. On fair use, it found three of the four statutory factors weighed against Ross. The use was commercial and, in the court's view, barely transformative, since both companies used the headnotes to build a legal research platform. Only the second factor, the largely factual nature of headnotes, favored Ross.
The court said copying was not necessary because Ross could have used the freely available court opinions themselves; it chose headnotes because they were easier. On market harm, the panel found Ross's tool competed with Westlaw and harmed Thomson Reuters's position in a potential market for licensing its headnotes as AI training data. Stripped of the AI framing, the court called this "no more than an ordinary copyright case."
A Thomson Reuters spokesperson said respecting copyright is essential to fostering innovation while protecting intellectual property, PYMNTS reported. Yar Chaikovsky of White & Case said on Ross's behalf that the decision leaves continued uncertainty about copyright law and AI training, and that Ross intends to seek Supreme Court review. Ross, which LawNext describes as shuttered, did not dispute in this appeal that the headnotes were copied on its behalf.
The numbers
- Westlaw headnotes at issue
- 2,243
- Training memos written by LegalEase
- About 25,000
- Judicial opinions in Ross's search bank
- About 10 million
- Fair use factors against Ross
- 3 of 4
- Share of Westlaw's 28 million headnotes Ross said it took
- 0.08% (Ross's figure)
Why CEOs should care
For general counsel and CTOs training models on content they did not create, the ruling closes one argument: an intermediate AI training step did not make copying fair when the end product competed with the original. The court also treated licensing content as AI training data as a potential market the owner can lose. Teams should map which training sets come from a competitor's proprietary material and whether a public-domain or licensed source could do the same job, since the court weighed that Ross had free alternatives.
For CFOs and owners of proprietary data, the decision strengthens the case that editorial and curated content is a licensable AI asset. Publishers, information services and software companies with annotated data may find their hand stronger in licensing talks. Buyers of AI tools should ask vendors where their training data came from, whether it was licensed and what indemnities cover a copyright claim.
Boards should not over-read the ruling. The panel stressed that Ross's system could not generate original expression and distinguished generative models. It also cited evidence that Ross at times acted in bad faith, including trying to access Westlaw with credentials it was told it could not use. How data is obtained, not only how it is used, is now part of the risk picture.
The bigger picture
The case is the first appellate test of AI training under fair use, but its facts are narrow. LawNext noted the copying happened before generative AI took off and involved a non-generative tool built to compete directly with Thomson Reuters. In a footnote, the panel pointed to a September 1, 2026, Justice Department statement of interest in the OpenAI copyright litigation in Manhattan federal court, which relied on Bartz v. Anthropic to argue that training models that generate original responses is transformative. The panel said those arguments did not apply to Ross.
What’s next
Ross has said it will ask the Supreme Court to review the decision. The panel ruled only on the two questions certified for interlocutory appeal, whether the headnotes are original and whether Ross's use was fair. For generative AI developers, the bigger tests remain the pending cases over large language models, where the Justice Department has argued that training models that generate original responses is transformative.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
