On November 13, 2026, Authors Alliance rounded up six amicus briefs supporting ROSS's bid for rehearing, putting Thomson Reuters v. ROSS back in the spotlight. It is the first U.S. appellate case to rule on whether training AI on copyrighted material is fair use: on September 29, 2026, a three-judge panel of the U.S. Court of Appeals for the Third Circuit held that 2,243 Westlaw headnotes are protected by copyright and that ROSS's use of them to train a legal-research AI was not fair use. ROSS filed its petition for rehearing en banc on November 3, and on November 9 copyright lawyer Heather Meeker filed another brief questioning whether the headnotes are copyrightable at all.
What the rehearing petition is fighting over
ROSS's petition takes aim at the panel's analysis of the first, third and fourth fair-use factors. On the first factor, it asks the court to separate the specific use — training a model on headnotes — from the companies' eventual commercial competition, rather than judging it by product rivalry alone. On the third, ROSS stresses that the headnotes never appeared in public outputs, so the amount copied should be measured by what was actually made public. On the fourth, it rejects treating 'a licence should have been paid for' as evidence of market harm, calling that circular reasoning. Meeker's brief goes to a more basic question: headnotes that accurately summarise public court decisions are short summaries constrained by fact and, under the merger doctrine, should not be protected by copyright; treating each headnote as a separate work could also multiply the base for statutory damages.
The six briefs do not all make the same case
Authors Alliance's November 13 roundup shows the six briefs divide the work. Authors Alliance argues that extracting only unprotected facts and ideas, without disseminating protected expression, is not infringement. Professor Brian Frye's brief notes that the panel cited no single copyrightable headnote as an example, making its conclusion hard to assess. Briefs from the Foundation for American Innovation and DingDuff focus on the fourth factor, arguing that the logic of 'usable for training plus no licence fee equals market harm' would hand rights holders the advantage in almost every case. Abraham Kang and Kunal Patel's brief deals specifically with the third factor. These briefs are only submissions asking the court to reconsider; on the public docket so far, the court has not ruled on the rehearing petition.
What to do now, and where the limits are
For teams training models on copyrighted material, the rehearing bid changes nothing by itself: the September 29 appellate ruling still stands, and the panel's conclusion continues to apply within the Third Circuit. The only thing worth watching is the court's next move — whether it grants rehearing and whether it accepts these briefs. Three limits matter for compliance. First, the case concerns a non-generative retrieval system, and the panel expressly distinguished generative-AI cases, so the ruling cannot be applied directly to training large language models. Second, the ruling binds only the Third Circuit; other circuits and pending cases elsewhere are not bound by it. Third, ROSS shut down back in 2021, so the outcome matters for how later comers assess the copyright risk of training data, not for a company still in business. Teams buying training data should still keep records of sources and licences, and adjust their approach once the court speaks.
