The Third Circuit finds that AI companies can’t use material with an original “creative spark.”

In October 2026, the U.S. Court of Appeals for the Third Circuit upheld a lower court decision in the case Thomson Reuters v. ROSS Intelligence Inc., ruling that AI companies cannot use copyrighted material with original creative elements for training or commercial purposes. The Third Circuit oversees Delaware, New Jersey, Pennsylvania, and the Virgin Islands. The case centered on ROSS Intelligence, an AI company that used legal research headnotes from Thomson Reuters' Westlaw platform to build a competing product. The court found that even though headnotes are summaries of judicial opinions, their arrangement and curation by Westlaw contained enough originality to be protected by copyright law. This decision reinforces that AI companies cannot claim fair use (a legal doctrine allowing limited use of copyrighted material without permission) when the material has a 'creative spark,' meaning it has been sufficiently transformed or organized to be considered original work.

Headnotes are short summaries placed at the top of judicial opinions, which outline the key legal issues and rulings in a case. These summaries are created by legal publishers like Thomson Reuters' Westlaw, which curate and arrange them to help lawyers and researchers quickly understand the significance of court decisions. While the underlying judicial opinions themselves are not copyrightable because they are facts, the way headnotes are written and organized can contain enough originality to qualify for copyright protection. In the Thomson Reuters v. ROSS case, the court determined that Westlaw's headnotes had a 'minimal creative spark' due to their selection, arrangement, and expression, making them legally distinct from the raw judicial opinions they summarize.

ROSS Intelligence argued that its use of Westlaw's headnotes fell under fair use, a legal doctrine that allows limited use of copyrighted material without permission for purposes like criticism, commentary, or education. However, the Third Circuit disagreed, stating that ROSS's use was 'minimally transformative at best' and primarily served to create a directly competing product. The court emphasized that fair use does not apply when the material used has originality and is being repurposed for a similar commercial purpose. This ruling is significant because it limits the ability of AI companies to claim fair use when training their systems on copyrighted works, even if the works are transformed or repackaged.

The Third Circuit's ruling has broader implications for how AI companies can use copyrighted material for training. The court's decision suggests that if a copyrighted work has been sufficiently transformed or curated to have a 'creative spark,' it cannot be used by AI companies without permission. This could complicate the process of building AI models, particularly in industries like publishing and legal research, where copyrighted texts are commonly used. The ruling may also encourage the development of new markets for 'laundering' media, where companies transform copyrighted works into something original that can be legally used for AI training. However, the decision is limited in scope, as it applies to a non-generative AI model and does not address the practices of major AI companies like Google, Meta, or Palantir.

Judge Tamika Montgomery-Reeves, who authored the Third Circuit's opinion, stated that the case is 'no more than an ordinary copyright case,' emphasizing that legal constraints still apply to AI companies despite their claims of technological disruption. The ruling highlights that copyright law remains a tool to regulate the AI industry, even as companies argue that AI requires special legal considerations. However, the decision is seen as a mixed bag, particularly for the publishing industry, as it blurs the line between facts and creative expression, expands copyright enforcement to intermediate copies, and could create new markets for AI training data. The case also underscores the challenges of defining fair use in the context of AI, where the transformation of copyrighted material is often subjective and open to interpretation.

The court's reliance on the concept of a 'creative spark' to determine copyrightability introduces a subjective element into the law, which some argue is problematic. The phrase echoes the Supreme Court's famous 'I know it when I see it' standard used in the 1964 case Jacobellis v. Ohio to define obscenity. Critics argue that such standards are vague and open to interpretation, potentially leading to inconsistent rulings. The article compares this legal measure to historical examples of subjective judgments, such as witchcraft trials, to illustrate the challenges of relying on intangible concepts like creativity in the law. This subjectivity may create uncertainty for AI companies and copyright holders alike, as the boundaries of what constitutes a 'creative spark' remain unclear.