ROSS Plans a Supreme Court Challenge to the Third Circuit’s Fair Use Ruling in Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc.

The debate over AI training licenses exposes the competition question left open by the copyright ruling.

ROSS Intelligence announced on October 2 that it intends to seek Supreme Court review of the Third Circuit’s decision in Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc. The court rejected fair use of Westlaw headnotes to train a competing legal research service. The planned challenge renews attention to a question with far-reaching consequences: how far copyright’s protection of potential training licenses should extend when the copyright owner also competes with the prospective licensee.[1]

Bob Ambrogi’s report for LawSites attributes the announcement to ROSS counsel Yar Chaikovsky. The company’s own post says it intends to seek review. The Supreme Court has not agreed to hear the case. The announcement nevertheless gives the substantial commentary on the ruling a new focus. Writers disagree about its implications for generative AI and the proper relationship between training, the resulting product, and the markets copyright should protect.[2]

In my earlier American Counsel article, I examined the power preserved by the ruling. A potential licensing market may receive copyright protection even when its owner has declined to license a competitor. That combination deserves scrutiny because it can influence who enters a market and what competing services become possible. The discussion following the decision helps identify where the legal disagreement ends and that policy inquiry begins.[3]

The underlying facts are relevant: ROSS sought to build a system that returned relevant passages from judicial opinions. Thomson Reuters declined to license its material to a competitor, so ROSS then obtained approximately 25,000 training memoranda through LegalEase Solutions, which themselves drew on Westlaw headnotes--Westlaw being a Thomson Reuters product/service. The appeal concerned partial summary judgment involving 2,243 headnotes that the court found sufficiently original for copyright protection. The judicial opinions themselves remained in the public domain.[4]

Judge Tamika Montgomery-Reeves’s opinion treated training as part of a commercial project serving substantially the same research purpose as Westlaw. The intermediate step did little to transform that purpose. The court also reasoned that ROSS could have worked from public judicial opinions without copying Westlaw’s protected expression. Of the four fair-use factors, only the nature of the copyrighted work favored ROSS. The court affirmed through an interlocutory appeal, leaving the litigation’s other questions outside this decision.[5]

The opinion identified more than one form of market harm. It considered the value of headnotes in attracting Westlaw subscribers, competition between research services, and the developing market for licensing headnotes as training data. Its licensing analysis rested partly on evidence of that developing market and Thomson Reuters’s use of headnotes in its own AI products. ROSS supplied no contrary evidence. Any criticism of the licensing rationale must account for those findings and the court’s separate substitution analysis.[6]

The initial coverage reflected the importance of an appellate decision in this field. Blake Brittain’s Reuters reportdescribed the affirmance while distinguishing ROSS from generative AI. Jackson Healy’s account for Courthouse News Service explains the similar-purpose and market-harm reasoning. Ambrogi’s analysis of the ruling likewise emphasizes the significance of a product built to compete directly with Westlaw.[7]

The brief interval before the opinion became public requires care in reading early reactions. Eric Hoffman at AI Esquireemphasized the limits of interpreting a sealed decision and the need to distinguish public law from proprietary editorial material. Steve Brachmann’s IPWatchdog report examined the trial court’s reasoning and reported criticism of its approach to intermediate copying. Those pieces document the first response to the outcome. The released opinion supplies the basis for assessing the appellate court’s reasoning.[8]

On X, Jason Kint welcomed the affirmance as a response to categorical assurances that AI training is fair use. Adam Eisgrau drew a different implication from the released opinion. He read its distinction between ROSS and generative systems as encouraging for highly transformative, general-purpose models. The disagreement concerns how to generalize from the decision. Kint emphasizes the failure of an AI training defense. Eisgrau emphasizes the characteristics that made this particular defense fail.[9]

Brandon Butler thinks ROSS should have won while also distinguishing its case from general-purpose AI litigation. The distinction has textual support. Footnote seven discusses ROSS’s nongenerative design and its effort to create a commercial substitute for Westlaw. It does not establish a rule that all generative training is fair. Nor does the result make every training use unlawful. The use’s purpose and its relationship to protected markets remain central to the analysis.[10]

The company resists attempts to confine the ruling’s importance. In an October 1 post, ROSS argues that training should be assessed separately from the resulting product and that the panel’s approach will affect other AI cases. That is a litigant’s argument about the decision’s reach. It identifies a substantive issue that any planned petition could develop: how courts should characterize copying undertaken during training when evaluating the purpose of that use.[11]

Evan Brown’s analysis at internetcases explains why the licensing question may persist even where the products differ. A decision grounded in direct competition can still furnish reasoning about training markets elsewhere. Brown identifies the circularity problem that arises when the possibility of a license becomes evidence that the use required one. Aaron Moss at Copyright Lately raised a related concern before the appellate reasoning was public while recognizing the force of the direct-substitution case against ROSS.[12]

The concern warrants a precise formulation. If every use for which an owner might demand payment establishes a protected market, the fair-use inquiry risks presupposing the permission requirement it is meant to examine. Courts therefore need a reason to recognize the asserted market beyond the owner’s desire to charge. Evidence of a developing market matters. So does the distinction between displacement of protected expression and competition made possible by a different use. The panel’s additional findings make this a more demanding critique than simply calling its licensing market hypothetical.

Catherine Zaller Rowland’s updated analysis summarizes the rationale supporting the copyright owner’s position. Common commercial purpose, competing research platforms, and the opportunity to license training material together informed the court’s decision.[13]

Editorial selection and expression require investment. A legal system that routinely allows competitors to appropriate that work could weaken the incentive to produce it. The difficulty lies in defining the protected opportunity without allowing the prospect of a licensing fee to determine the fair-use outcome by itself.

The music commentary shows why that difficulty extends beyond legal research. Murray Stassen at Music Business Worldwide explains the industry’s interest in substitution and emerging licensing markets. Brian McBrearty at Musicologize examines the distinction between copying necessary to accomplish a new use and copying that makes development more convenient. Dylan Smith at Digital Music News considers the implications for music AI and licensing value. Each recognizes that the differences between ROSS and generative products matter. A recurring economic argument can influence later cases even when their particular facts require different outcomes.[14]

My concern begins with the resulting allocation of power. The recognition of a licensing market does not itself require licenses to be offered. An incumbent may preserve the value of editorial work while retaining discretion over access by firms that might compete with it. That arrangement can be lawful under the copyright decision and still warrant an inquiry into its effects on entry. The appellate court did not resolve an antitrust claim or establish that a refusal to license must be remedied. A policy analysis should examine the costs of independent development and the practical availability of alternative inputs before prescribing an access obligation.[15]

Public access to judicial opinions provides an important starting point. The panel treated those opinions as an available alternative to copying headnotes. Policymakers should examine how usable that alternative is for new research services. Investment in accessible, reusable public legal materials could reduce entry costs while preserving protection for original editorial contributions. Any proposal to compel licensing would require its own legal justification and evidence about effects on both competition and investment. Copyright’s boundaries and the practical conditions of market entry deserve separate attention.

The October 2 update at Chat GPT Is Eating the World identifies the Solicitor General’s possible role as a development to watch. The opinion itself acknowledges a government argument made in separate generative-AI litigation. That reference does not establish a United States position on a future ROSS petition. The blog’s expectation of government involvement remains a prediction.[16]

A petition would clarify which questions ROSS actually asks the Supreme Court to decide. Until then, the most useful contribution is to identify what would justify treating a training market as protected and what evidence should connect training to market harm. The policy inquiry should proceed alongside that doctrinal work. The ability to protect editorial investment is valuable. Whether the resulting control leaves workable opportunities for competing legal research services requires an examination of how those services can actually be built.

---

[1] Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., No. 25-2153, slip op. at 7, 10–13, 24–28 (3d Cir. Sept. 29, 2026) (affirming partial summary judgment on originality and fair use).

[2] Bob Ambrogi, ROSS Says It Will Ask Supreme Court to Review 3rd Circuit Ruling for Thomson Reuters in Copyright Case, LawSites (Oct. 2, 2026) (reporting counsel’s announcement of intended review). ROSS (@ROSSIntel), Post, X (Oct. 2, 2026) (announcing an intention to seek review).

[3] J.R. Howell, The Third Circuit’s Decision in Ross Protects Westlaw’s Market for Licensing Headnotes to AI Rivals. Now what?, The American Counsel (Oct. 2, 2026) (examining the competition implications of protection for a potential licensing market).

[4] Thomson Reuters, slip op. at 3–7, 10–13 (describing the training memoranda and affirming originality findings concerning 2,243 headnotes).

[5] Thomson Reuters, slip op. at 7, 13–24, 28 (explaining the interlocutory posture and weighing the fair-use factors).

[6] Thomson Reuters, slip op. at 24–27 (considering subscription value, competition, and evidence of a developing training-data licensing market).

[7] Blake Brittain, US Appeals Court Upholds Thomson Reuters’ Landmark Win in AI Training Lawsuit, Reuters (Sept. 29, 2026), republished by MarketScreener (reporting the affirmance before the appellate reasoning became public). Jackson Healy, AI Training of Copyrighted Material Not Fair Use: Third Circuit, Courthouse News Service (Sept. 30, 2026) (describing the court’s reasoning and the parties’ reactions). Bob Ambrogi, 3rd Circuit Rules for Thomson Reuters in Its Copyright Fight Against Legal Research Startup ROSS, LawSites (Sept. 30, 2026) (explaining the significance of ROSS’s competing nongenerative product).

[8] Eric Hoffman, The First Appellate AI Copyright Ruling Arrived Under Seal, AI Esquire (Sept. 30, 2026) (distinguishing public legal materials from editorial contributions and discussing the initial absence of public reasoning). Steve Brachmann, Third Circuit Affirms Revised Fair Use Ruling Against ROSS’ AI Legal Research Platform in Sealed Opinion, IPWatchdog (Sept. 30, 2026) (reviewing the lower court’s approach and reporting criticism from Brandon Butler).

[9] Jason Kint (@jason_kint), Post, X (Sept. 29, 2026) (challenging categorical claims about fair use in AI training). Adam Eisgrau (@AdamEisgrau), Thread, X (Sept. 30, 2026) (arguing that the opinion’s reasoning leaves room for fair use by transformative general-purpose models).

[10] Brandon Butler (@bc_butler), Post, X (Sept. 30, 2026) (disagreeing with the result while distinguishing general-purpose AI cases). Thomson Reuters, slip op. at 5, 17 n.7 (distinguishing ROSS’s product and purpose from arguments concerning generative systems).

[11] ROSS (@ROSSIntel), Post, X (Oct. 1, 2026) (arguing that the ruling’s treatment of training will affect other AI cases and quoting Andrew Arruda’s criticism of the input-and-output analysis).

[12] Evan Brown, The First Appellate AI Training Decision Is Not Really a Generative AI Case, internetcases (Oct. 1, 2026) (examining the potential reach of the licensing-market rationale and the risk of circularity). Aaron Moss, Signed, Sealed, Affirmed: Third Circuit Backs Thomson Reuters over ROSS, Copyright Lately (Sept. 29, 2026) (questioning aspects of the trial court’s reasoning before the appellate opinion became public).

[13] Catherine Zaller Rowland, Post, LinkedIn (edited post reviewed Oct. 2, 2026) (explaining the released opinion’s analysis of common purpose, competition, and licensing opportunities).

[14] Murray Stassen, US Appeals Court Rejects ‘Fair Use’ Defense over AI Training – as Thomson Reuters Wins Copyright Case Backed by RIAA and NMPA, Music Business Worldwide (Sept. 30, 2026) (describing music-industry arguments while distinguishing pending generative-AI cases). Brian McBrearty, Is AI Training Fair Use? The First Appeals Court to Answer Just Said Nah, Musicologize (Sept. 30, 2026) (examining necessity, convenience, and substitution in training). Dylan Smith, Thomson Reuters Scores Appeals Court Win In Closely-Watched ‘Fair Use’ Copyright Case — Defendant’s Copying Was ‘Highly Commercial and Minimally Transformative’, Digital Music News (Sept. 30, 2026) (discussing possible implications for music AI and licensing value).

[15] Thomson Reuters, slip op. at 4, 7, 19–21, 24–28 (describing the declined license and resolving the certified copyright questions). Howell, supranote 3 (discussing the distinction between the copyright ruling and questions about competition and access).

[16] ROSS Intelligence to Seek Supreme Court Review of Third Circuit’s Decision Rejecting Fair Use. All Eyes Will Be on SG’s Position., Chat GPT Is Eating the World (Oct. 2, 2026) (predicting possible Solicitor General involvement). Thomson Reuters, slip op. at 17 n.7 (addressing a government argument advanced in separate generative-AI litigation).