{"id":9849,"date":"2026-10-10T03:55:49","date_gmt":"2026-10-10T03:55:49","guid":{"rendered":"https:\/\/journals.law.unc.edu\/ncjolt\/?p=9849"},"modified":"2026-10-10T03:56:18","modified_gmt":"2026-10-10T03:56:18","slug":"headnotes-not-headlines-what-the-third-circuits-ross-decision-actually-settles-about-ai-training","status":"publish","type":"post","link":"https:\/\/journals.law.unc.edu\/ncjolt\/blogs\/headnotes-not-headlines-what-the-third-circuits-ross-decision-actually-settles-about-ai-training\/","title":{"rendered":"Headnotes, Not Headlines: What the Third Circuit\u2019s ROSS Decision Actually Settles About AI Training"},"content":{"rendered":"\n

12:01AM, October 10, 2026<\/strong><\/p>\n\n\n\n

\"\"
Credit: https:\/\/pixabay.com\/photos\/gavel-justice-judge-courtroom-7499911<\/figcaption><\/figure><\/div>\n\n\n\n

In\u00a0Thomson Reuters v. ROSS Intelligence<\/a>, decided by the Third Circuit on September 29, 2026, the court became the\u00a0first federal appellate court<\/a>\u00a0to rule on whether using copyrighted material to train an artificial intelligence program\u00a0constitutes\u00a0fair use. For ROSS\u2019s use, the court said no. No sooner had the ruling been issued than media headlines<\/a>\u00a0described a first appellate loss for AI training. A key limit appears in\u00a0a footnote\u00a0of the decision.\u00a0<\/p>\n\n\n\n

By technology standards, these facts feel ancient. ROSS aimed at developing a legal search engine that would respond to a question posed in natural language by quoting passages from judicial opinions.\u00a0To train this engine, ROSS needed some examples of legal questions together with appropriate answers.\u00a0After Thomson Reuters refused to license its content to ROSS because it was a competitor, ROSS hired a contractor to prepare approximately 25,000 \u201cbulk memos,\u201d question-and-answer sets created using Westlaw headnotes, that is, summaries of legal points of a judicial opinion written by the editors. Thomson Reuters sued. In\u00a0February 2025<\/a>, Judge Stephanos Bibas (sitting by designation in the District of Delaware) granted partial summary judgment concerning 2,243 headnotes and rejected ROSS\u2019s fair-use defense.\u00a0<\/p>\n\n\n\n

Judge Montgomery-Reeves, writing for a unanimous panel,\u00a0affirmed<\/a>. Headnotes have \u201csome creative spark,\u201d because choosing which points of law to extract and how to phrase them involves creative judgment. The merger doctrine cannot help ROSS since \u201cjust as a banana costume can take myriad designs, so too can headnotes have differing expressions.\u201d\u00a0<\/p>\n\n\n\n

In terms of the first factor, both companies used the headnotes in creating their tools for legal research, making ROSS\u2019s use of headnotes \u201cminimally transformative, at best.\u201d In the\u00a0intermediate-copying cases<\/a>\u00a0dealing with software (Sega, Connectix, Google v. Oracle), ROSS\u2019s case was not strengthened since the copying in those instances was needed to get to the unprotected functional components of the computer code. Moreover, ROSS could have used\u00a0the\u00a0opinions. However, according to the court, it chose to copy the headnotes because that offered\u00a0an easy way\u00a0to create training memos. In the fourth factor, ROSS competed with the current market for legal research and usurped Thomson Reuters\u2019s opportunity to enter a developing market for licensing headnotes as AI training data. The judgment was affirmed, although not\u00a0all of\u00a0the reasoning used: Judge Bibas had ruled in ROSS\u2019s favor on the third factor, and the panel disagreed on that factor.\u00a0<\/p>\n\n\n\n

The first appellate word on AI training turned out to be about a search engine. The question the headlines were asking, what happens when the model writes, is still waiting for its appellate court.<\/p><\/blockquote>\n\n\n\n

Now the footnote.\u00a0On September 1, 2026, the Department of Justice filed a\u00a0statement of interest<\/a>\u00a0in the OpenAI litigation in New York,\u00a0telling\u00a0that court that training a large language model is transformative because the model goes on to produce new responses of its own. The Third Circuit did not say the government was wrong. It said the argument does not fit ROSS: unlike the models in Bartz and\u00a0In\u00a0re: OpenAI, ROSS\u2019s \u201cAI platform cannot generate original expression,\u201d and its tool did the same job as Westlaw. With that, the court set aside the questions that make the generative cases hard. Is a model that learns from millions of books\u00a0in order to\u00a0write new text doing something transformative? Is a flood of machine-written output the kind of market harm copyright recognizes? Both\u00a0remain\u00a0open.\u00a0<\/p>\n\n\n\n

This is significant because the trial courts have taken different approaches to generative AI training. In June 2025, Judge William Alsup in\u00a0Bartz v. Anthropic<\/a>, sitting in the Northern District of California, concluded that training Claude on books was \u201cexceedingly transformative,\u201d like someone who reads with the intent of becoming an author, while also concluding that downloading pirated books to build a permanent library was not fair use. It is the latter conclusion that led to the\u00a0$1.5 billion\u00a0settlement<\/a>\u00a0approved in July 2026. Two days after Alsup\u2019s decision, Judge Vince Chhabria issued summary judgment for Meta in\u00a0Kadrey v. Meta<\/a>, but only \u201con this record,\u201d noting that \u201cin many circumstances it will be illegal to copy copyright-protected works to train generative AI models without permission\u201d due to their ability to cause market dilution. All three cases illustrate different views of transformation and market effects, with both generative AI cases finding training fair use, while the appellate ruling\u00a0concerned\u00a0a search engine.\u00a0<\/p>\n\n\n\n

Still, parts of the reasoning will remain relevant. Firstly, \u201cAI training\u201d is not a magic phrase. The court considered the function of the tool rather than the means used to create it, and a defendant who created a similar\u00a0end product\u00a0will struggle on factor one regardless of the technology. Secondly, necessity has significance. Where the unprotected element (the opinions) is accessible through a means other than reproduction of the protected element (the headnotes), the intermediate-copying doctrine will be of no help to the defendant. Thirdly, and most importantly for the plaintiffs, the court recognized a potential market for licensing Westlaw headnotes for AI training and treated ROSS\u2019s copying as harming it. Such reasoning has been\u00a0criticized<\/a>\u00a0as circular: copyright owners can always claim they would have charged for the disputed use. Whether circular or not, it\u00a0remains\u00a0the law in the Third Circuit, and a valuable argument for the plaintiffs in the generative cases, even if its reach beyond headnotes is contested.\u00a0<\/p>\n\n\n\n

The first appellate word on AI training\u00a0turned\u00a0out to be about a search engine. The question the headlines were asking, what happens when the model writes, is still waiting for its appellate court.\u00a0<\/p>\n\n\n\n

Esther Dong<\/strong><\/p>\n\n\n\n

Esther Dong is a 2L at the University of North Carolina School of Law and a staff member of the North Carolina Journal of Law & Technology. Esther graduated with highest distinction from UNC Chapel Hill with a B.A. in Communication Studies and is interested in technology transactions, intellectual property, and artificial intelligence law. When not in class, you can find her playing tennis, hiking, or being bossed around by her cat, Onyx.<\/p>\n","protected":false},"excerpt":{"rendered":"

The rule had been shown to reduce EtO emissions by more than 90%, accordingly reducing cancer risk by 92%. Despite additional findings that the gas is 60 times more toxic than previously thought, the new EPA proposed rule plans to roll back these standards.<\/p>\n","protected":false},"author":4,"featured_media":9768,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[51],"tags":[722,724,725,509,723],"_links":{"self":[{"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/posts\/9849"}],"collection":[{"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/comments?post=9849"}],"version-history":[{"count":2,"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/posts\/9849\/revisions"}],"predecessor-version":[{"id":9852,"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/posts\/9849\/revisions\/9852"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/media\/9768"}],"wp:attachment":[{"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/media?parent=9849"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/categories?post=9849"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/journals.law.unc.edu\/ncjolt\/wp-json\/wp\/v2\/tags?post=9849"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}