Third Circuit says curated text keeps its copyright against AI

A court just told an AI company that "we transformed it" isn't a magic password. That's the headline, and if you write or publish anything with structure, curation, or editorial judgment baked into it, you should care.

Here's what happened. Thomson Reuters sued ROSS Intelligence back in 2020, claiming ROSS had ripped off the headnotes — those short summaries sitting atop judicial opinions — from Reuters's Westlaw platform to build a competing legal research tool. ROSS argued fair use, the defense that's worked for AI companies before. In 2025, a judge disagreed: even though Westlaw's headnotes draw on judicial opinions that aren't themselves copyrightable, Westlaw's arrangement and curation of that material had what the ruling called "the minimum 'spark' of originality." That was enough to make it protected work.

Last week, the Third Circuit — covering Delaware, New Jersey, Pennsylvania, and the Virgin Islands — upheld that finding on appeal. U.S. Circuit Judge Tamika Montgomery-Reeves ruled Westlaw's headnotes carry "minimal 'creative spark,'" and that ROSS's use of them was "minimally transformative at best" in service of a directly competing product. Translation: illegal.

Here's your personal stake. If your work involves selecting, arranging, or editorially shaping material — even material built from facts or public-domain sources — this ruling gives you a legal hook. Courts are willing to call that "spark" real and protectable, and willing to call copying it for a competing AI product exactly what it is: infringement, not transformation.

But don't pop the champagne yet. Judge Montgomery-Reeves herself wrote that despite appearances, "this is no more than an ordinary copyright case" — not a sweeping verdict on generative AI. And ROSS was a minor player, already out of business. The bigger worry, raised by Authors Alliance's Yuanxiao Xu, is that this ruling could cut two ways: it blurs the line between fact and expression, expands copyright claims to intermediate copies, and — more unsettling — it might hand bigger AI companies a map for how to "launder" data into something that looks transformed enough to use. The giants doing the real damage to the book trade aren't ROSS-sized. They're Google, Meta, Palantir scale.

What to actually do this week: if you or your publisher license curated databases, anthologies, annotated editions, or any product built on arrangement and selection, flag this ruling to your rights department now. It strengthens your hand in any dispute over AI scraping of that material. And keep watching for how this reasoning gets cited in the bigger pending cases against the larger AI players — that's where you'll find out if "creative spark" is a real shield or just a nice phrase from a small case.