Appeals Court Weighs AI Training on Legal Research
Ross Intelligence's rehearing request tests whether training on copyrighted headnotes can qualify as fair use under current law.
Updated on Nov. 3, 2026 in Artificial Intelligence
Live Poll
Should AI companies be allowed to train their tools on copyrighted material without permission?
Ross Intelligence has asked for a new appellate hearing in its copyright dispute with Thomson Reuters, after a three-judge Third Circuit panel ruled against its fair use defense in September. The AI legal research company, which trained its model on Westlaw headnotes, filed the request on Wednesday. This is an active legal dispute, not a shipped product change.
Why it matters
The case tests whether training an AI system on copyrighted legal research material can qualify as fair use, a question with direct consequences for how AI developers source training data. The Third Circuit's holding that Thomson Reuters's materials are copyrightable and that Ross's use was not fair sets a significant appellate marker in a field where most training-data disputes remain unresolved.
The dispute centers on Ross's use of Westlaw headnotes, short summaries of legal points, to produce legal memoranda that trained its AI model. The court record documents the litigation timeline from the 2020 filing through the February 2025 district ruling and the September appellate decision, but the stub does not include benchmark or performance figures for the Ross model.
The players
Ross Intelligence
An AI legal research startup whose platform was trained on Westlaw headnotes and which is now seeking a rehearing of its copyright loss.
Thomson Reuters
The information services company behind Westlaw, whose legal research database and editorial headnotes are the copyrighted material at the center of the case.
Judge Stephanos Bibas
The federal judge who ruled in February 2025 that Ross was not entitled to a fair-use defense.
NetChoice
A trade group representing internet and technology companies that filed a friend-of-the-court brief supporting Ross.
The details
Ross Intelligence built an AI legal research tool and used Westlaw headnotes, the summaries Thomson Reuters writes to describe points of law in court opinions, to generate legal memoranda that served as training material. Thomson Reuters alleged that Ross obtained the headnotes through a licensee and used them to bring a competing product to market faster than it could have built one itself. The Third Circuit held that the headnotes are copyrightable and that Ross's use did not qualify as fair use, while Entertainment and publishing groups backed Westlaw and NetChoice backed Ross in friend-of-the-court briefs.
Timeline
2020: Thomson Reuters sued Ross over its use of Westlaw headnotes.
February 2025: Judge Stephanos Bibas ruled Ross was not entitled to a fair-use defense.
September: A three-judge Third Circuit panel ruled against Ross.
Wednesday: Ross filed its request for a new appellate hearing.
The Tech Race
The case places AI legal research tools in direct conflict with the publishers whose editorial work supplies their training material. The outcome of Ross's rehearing request will shape how other AI developers approach licensing and fair-use claims for training data.
Developers of AI tools that rely on licensed or publicly available legal and editorial material could face changed licensing costs depending on how the Third Circuit handles the rehearing request. The stub does not provide a timeline for a decision on the request, so the practical effect on products is not yet known.
The takeaway
Watch whether the Third Circuit grants Ross's request for a new appellate hearing, which would determine whether the September ruling stands. The court's decision on that request is the next verifiable milestone for how AI training on copyrighted legal research will be treated.
Further reading
For broader context on how courts are treating AI training data, see the Artificial Intelligence coverage.
Source note: This article includes information reported by MediaPost.
Live Poll
Should AI companies be allowed to train their tools on copyrighted material without permission?
