Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc

UNITED STATES COURT OF APPEALS FOR THE THIRD CIRCUIT No. 25-2153 THOMSON REUTERS ENTERPRISE CENTRE GMBH; WEST PUBLISHING CORP v. ROSS INTELLIGENCE INC, Appellant _____________________________ On Appeal from the U.S. District Court, D. Del. Judge Stephanos Bibas,* No. 1:20-cv-00613 Before: RESTREPO, MONTGOMERY-REEVES, and BOVE, Circuit Judges Argued: June 11, 2026; Filed: Sept. 29, 2026 _____________________________ OPINION OF THE COURT MONTGOMERY-REEVES, Circuit Judge. Artificial intelligence (“AI”) is a powerful machine learn- ing technology that is poised to impact many facets of Ameri- can life, including how we work, how we make art, and, criti- cally for this case, how we study the law. In this case, the * The Honorable Stephanos Bibas, Circuit Judge sitting by des- ignation pursuant to 28 U.S.C. § 291(b). veteran creator of a legal staple, Thomson Reuters Enterprise Center GmbH (“Thomson Reuters”), charges a new competitor with copying its editorial materials to train an AI platform. The newcomer, ROSS Intelligence, Inc. (“ROSS”), counters that Thomson Reuters seeks to squelch innovative AI technology. But new technology is often at the forefront of our copyright cases. And when it is, we regularly apply “the principles un- derlying our venerable copyright laws.” Whelan Assocs., Inc. v. Jaslow Dental Lab’y , Inc., 797 F.2d 1222, 1224 (3d Cir. 1986). Stripped to its essence, this interlocutory appeal asks us to do just that. Under ROSS’s framing, this case appears to concern the future of AI legal technology. But appearances can be deceiv- ing. In truth, this is no more than an ordinary copyright case; we must determine whether Thomson Reuters’s materials are sufficiently original for copyright protection and whether ROSS’s use of those materials was fair. The undisputed evi- dence reflects that Thomson Reuters’s materials possess a cre- ative spark, and ROSS aspired to be a direct competitor by us- ing them for a highly similar purpose. We therefore hold that Thomson Reuters’s materials are copyrightable and that ROSS’s use was not fair. Thus, we will affirm the District Court’s order granting Thomson Reuters partial summary judgment. I. FACTS & PROCEDURAL HISTORY This story begins with our country’s founding. As the United States took its first steps as an independent nation, it 2 began developing its own common law. In these early days, there were few official reporters publishing this developing corpus of American case law. Private commercial publishers sought to fill this void, and case reporters proliferated. One such publisher was West Publishing Corporation, which exists today as part of Thomson Reuters. Later, Thomson Reuters created Westlaw, an online legal- research platform that collects, organizes, annotates, and pub- lishes legal content such as judicial opinions. Westlaw organ- izes judicial opinions using its topic and Key Number System, which indexes opinions based on the legal topics addressed therein. These topics are connected to and expressed in head- notes. Headnotes are annotations on opinions that explain “what the law is in the abstract or the result which it compels on the facts of the case.” Appendix (hereinafter App. __”) 6510. In Westlaw’s user interface, the headnotes sit above the opinion, and clicking on a headnote takes the Westlaw user to the corresponding passage of the opinion upon which the head- note is based. Headnotes reflect a point of law made in the judicial opinion and are “distinguished from statements of fact and from [statements explaining the] reasoning [supporting the point of law].” App. 6510. When drafting headnotes, Thomson Reuters’s editors must follow precise guidelines. They must make each headnote clear and concise, including only the essential facts for the le- gal contention or holding at issue. They are directed to limit each headnote to 800 characters, where possible. Headnotes 3 are meant to convey to a reader the “most important points” in an opinion. App. 8219. Accurate description of the law is par- amount for headnotes, so the editors “[g]enerally . . . follow the court’s language . . . [but] each headnote must have enough context to stand on its own, and make sense to a reader who doesn’t have the opinion.” App. 8233. Thomson Reuters owns a registered copyright in Westlaw and its content, including the Key Number System, headnotes, and synopses at the beginning of opinions. Thomson Reuters disclaims any copyright in “any part of the original work pre- pared by a United States Government officer or employee as part of that person’s official duties,” such as the opinions un- derlying the headnotes. App. 176. Thomson Reuters also used its headnotes to train other Westlaw products like WestSearch Plus, an AI powered search function. But Westlaw has competitors in the legal-research industry, including ROSS. ROSS was founded by three University of Toronto computer science students who entered an IBM com- petition, in which competing teams used IBM’s Watson com- puter to solve a data-related problem in the team’s chosen in- dustry. The trio’s idea was to use Watson to train a computer program to answer legal research questions without using tra- ditional Boolean or keyword searches. After this competition, 4 the students founded ROSS to expand their idea and build an AI legal-research platform.1 ROSS then created an AI legal search engine that would respond to plain-language legal questions with relevant pas- sages of text from judicial opinions. ROSS had access to a bank of approximately ten million uncopyrighted judicial opin- ions from which it could respond with an excerpt relevant to a user’s query. ROSS’s AI was not a generative AI, meaning it would not create any new expression; it would only return text passages from preexisting judicial opinions. For ROSS’s AI legal search engine to work, the AI program had to learn what made a judicial opinion responsive to a user’s legal question. To train its AI program to do so, ROSS enlisted a company called LegalEase Solutions (“LegalEase”) to create AI training memoranda. LegalEase wrote approximately 25,000 legal memos, each of which presented a legal question and four to six judicial opinion passages in response. Memo- drafters for LegalEase, and its subcontractor Morae Global, used thousands of Westlaw headnotes to create the memos’ 1 AI here refers to machine learning technology in which a soft- ware program analyzes data to learn how to perform a task. The program is given data which contains examples of inputs paired with outputs. By analyzing enough data, the program can improve its performance, effectively learning to recognize patterns regarding which outputs are responsive to which in- puts. 5 legal questions, as the headnotes provided “an easy way [to] fram[e] questions.” App. 8545. The memos’ responses to the legal questions needed to include at least one great, good, top- ical, and irrelevant judicial opinion passage, labeled as such. A great passage was completely responsive; a good passage was mostly responsive; a topical passage was somewhat re- sponsive; and an irrelevant passage was not responsive. The “[g]reat” passages were “[m]ost frequently” the judicial opin- ion passages that Westlaw linked to the headnote that ROSS used as the memo’s legal question.2 App. 4837–39. These memos were then converted into a machine-readable format and used to train the ROSS AI program to identify patterns re- garding which judicial opinion passages respond well to legal questions. ROSS did not hide its intention that this for-profit AI plat- form compete with Westlaw. ROSS ran ads directly compar- ing its platform to Westlaw at prices “in line with” Westlaw. App. 4519. ROSS executives hoped ROSS would serve as a substitute for Westlaw, and some law firms switched from us- ing Westlaw to ROSS. Upon learning of ROSS’s actions, Thomson Reuters sued ROSS, alleging copyright infringement and tortious 2 For purposes of this appeal, ROSS is not challenging (1) that LegalEase and Morae Global’s actions are attributable to ROSS and (2) that LegalEase and Morae Global copied Westlaw headnotes. 6 interference with contract. The District Court ultimately granted partial summary judgment for Thomson Reuters, hold- ing as a matter of law that ROSS’s use of 2,243 Westlaw head- notes3 infringed Thomson Reuters’s copyright and was not fair use. Then, at the parties’ request, the District Court certified for interlocutory review “(1) whether the West headnotes and the West Key Number System[4] are original as a matter of law and (2) whether ROSS’s alleged use of the headnotes was fair use.” App. 61. II. JURISDICTION AND STANDARD OF REVIEW The District Court had jurisdiction under 28 U.S.C. §§ 1331, 1338. We have jurisdiction to hear this interlocutory ap- peal under 28 U.S.C. § 1292(b). We are “generally constrained 3 An expert for ROSS stated that the memo questions corre- sponding to a batch of 2,830 headnotes closely resembled the headnote text but differed significantly from the underlying ju- dicial opinions’ text. The District Court scrutinized this batch, comparing the headnotes to the memo questions and the opin- ions. The District Court concluded that for 2,243 headnotes, the memo questions were so similar to the headnote text but so dissimilar to the opinion text that no reasonable juror could conclude these headnotes were not copied. 4 ROSS never challenged the originality of the Key Number System in any of its briefing, despite its inclusion in the certi- fied questions. We do not address that forfeited issue. 7 to the questions certified for review” in an interlocutory appeal, but we “may consider any grounds justifying reversal.” Ellis v. Westinghouse Elec. Co., 11 F.4th 221, 229 (3d Cir. 2021) (quoting Morris v. Hoffa, 361 F.3d 177, 196 (3d Cir. 2004)). We review a grant of summary judgment de novo. TD Bank N.A. v. Hill, 928 F.3d 259, 270 (3d Cir. 2019). Summary judg- ment is appropriate when “there is no genuine dispute as to any material fact and the movant is entitled to judgment as a matter of law.” Fed. R. Civ. P. 56(a). “We view the facts in the light most favorable to the non-moving party and draw all reasona- ble inferences in that party’s favor.” Tundo v. Cnty. of Passaic, 923 F.3d 283, 287 (3d Cir. 2019) (citation modified). III. ANALYSIS Our Constitution empowers Congress “to promote the pro- gress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries.” U.S. Const., Art. 1, § 8, cl. 8 (cita- tion modified). Congress promotes the progress of science and arts through the Copyright Act, 17 U.S.C. § 101 et seq., which “encourage[s] creativity by granting to the author of an original work ‘a bundle of exclusive rights,’” like the rights to “repro- duce” a copyrighted work and “prepare derivative works.” Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 526 (2023) (quoting Harper & Row, Publishers, Inc. v. Nation Enters., 471 U.S. 539, 546 (1985)). 8  But the Copyright Act does not focus solely on encouraging creativity. It also promotes the availability of creative works. Stated differently, it “reflects a balance of competing claims upon the public interest: Creative work is to be encouraged and rewarded, but private motivation must ultimately serve the cause of promoting broad public availability of” creative works. Id. (quoting Twentieth Century Music Corp. v. Aiken, 422 U.S. 151, 156 (1975)). That “balancing act between crea- tivity and availability (including for use in new works)” is re- flected, in part, in the “fair use” defense, a common-law doc- trine codified at 17 U.S.C. § 107. Id. at 526–27. To establish a claim of copyright infringement, one must demonstrate “ownership of a valid copyright and copying of original elements of the plaintiff's work.” Am. Soc'y for Test- ing & Materials v. UpCodes, Inc., 172 F.4th 253, 261 (3d Cir. 2026) (citing Whelan Assocs., 797 F.2d at 1231). But one can have a copyright only in “original works of authorship.” 17 U.S.C. § 102(a). Even if these elements are satisfied, the “fair use” of a work’s original elements “for purposes such as criti- cism, comment, news reporting, teaching . . ., scholarship, or research, is not an infringement of copyright.” Id. § 107. This case turns on the originality requirement and ROSS’s fair use defense. We first consider whether Thomson Reuters’s head- notes were original, and then whether ROSS’s copying them into AI training memos for its legal-research platform was a fair use. 9  A. Originality5 We begin with originality. “Copyright protection sub- sists . . . in original works of authorship.” 17 U.S.C. § 102(a). This is an “extremely low” bar, as works need only be “inde- pendently created” with a “modicum of creativity” for copy- right protection; such creativity is present when a work “pos- sess[es] some creative spark.” Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340, 345–46 (1991). Although “the vast majority of [works] will pass this test,” “[t]here remains a nar- row category of works in which the creative spark is utterly lacking or so trivial as to be virtually nonexistent.” Id. at 359 (holding that the alphabetical arrangement of a phonebook does not possess the requisite “creative spark”). All 2,243 independently created headnotes at issue in this appeal have “some creative spark.” Id. at 345. For each head- note, Thomson Reuters’s editors made creative decisions about which points of law were important enough to include in a headnote and how to word those points of law consistent with Westlaw’s drafting criteria. Each headnote had to convey enough factual background and reasoning to stand on its own 5 The District Court mused that even headnotes that quoted ju- dicial opinions verbatim may be original enough for copyright protection. While this dictum raises an interesting question, we do not answer it here as the District Court only granted summary judgment as to 2,243 headnotes that do not copy opinion text verbatim. 10 without reference to the opinion or any other headnote, while accurately reflecting the corresponding opinion passage. Be- cause these decisions show the requisite minimal “creative spark,” the headnotes are original. Id. Supreme Court authority supports this conclusion. Indeed, the Supreme Court has suggested that headnotes prepared by private case reporters like Thomson Reuters are copyrightable. In Callaghan v. Myers, 128 U.S. 617 (1888), the Court held that works produced by judges in their official capacity were not copyrightable, but a reporter may obtain “a copyright for the volume which will cover the matter which is the result of his intellectual labor” such as “head-notes.” Id. at 647–49. Over a century later, the Court applied this principle to work produced by legislators in their legislative capacity, noting that Callaghan “upheld the reporter’s copyright interest in several explanatory materials that the reporter had created himself: headnotes, syllabi, tables of contents, and the like” so long as “they came from an author who had no authority to speak with the force of law.” Georgia v. Public.Resource.Org, Inc., 590 U.S. 255, 265 (2020) (emphasis added). ROSS offers three counters. First, ROSS argues that rec- ognizing Thomson Reuters’s copyright in its headnotes “would effectively give [Thomson Reuters] a monopoly over the law.” Opening Br. 22. Not so. Headnotes are not law; judicial opin- ions are. And judicial opinions remain “free for publication to all.” Banks v. Manchester, 128 U.S. 244, 253 (1888). 11  Second, ROSS argues the headnotes are not copyrightable under the merger doctrine. This doctrine precludes copyright in the “rare” case where “a work’s underlying idea can effec- tively be expressed in only one” or a “few . . . ways” such that “copyrighting a design feature would effectively monopolize an underlying idea.” Silvertop Assocs. Inc. v. Kangaroo Mfg. Inc., 931 F.3d 215, 222 (3d Cir. 2019) (quoting Educ. Testing Servs. v. Katzman, 793 F.2d 533, 539 (3d Cir. 1986)). But the merger doctrine does not apply when there “are many other ways to” convey a work’s underlying idea. Id. at 223. So, for instance, we have said that the merger doctrine does not apply to a “banana costume” due to the possibility of different cos- tumes based on “shape, curvature, tips, tips’ color, overall color, length, width, lining, texture, and material.” Id. And just as a banana costume can take myriad designs, so too can headnotes have differing expressions. Indeed, copyrighting the headnotes does not effectively monopolize judicial opinions because, as the District Court correctly noted, “there are many ways to express points of law from judicial opinions.” App. 49; see, e.g., Amicus Br. for RELX, Inc. (hereinafter the “Lexis Br. ___”) at 13–16. So the merger doctrine does not apply. Third and finally, ROSS cites Matthew Bender & Co. v. West Publishing Co., 158 F.3d 674 (2d Cir. 1998), to support the conclusion that headnotes are insufficiently original for copyright protection. But that out-of-circuit case does not sup- port ROSS’s argument. The Second Circuit held that West lacked a copyright in its arrangement of information like party names and its addition of information like parallel citations 12 because that information was not sufficiently original. Id. at 677. Those materials lacked originality, according to the court, because “industry conventions . . . dictate[d]” the selections and because “creativity inheres in making non-obvious choices from among more than a few options.” Id. at 682. In contrast, the Second Circuit described headnotes as “independently composed,” and it never suggested that industry standards dic- tated headnote composition. Id. at 676. Indeed, such reasoning would make little sense because Thomson Reuters’s editors make independent judgments about which points of law to in- clude and how to concisely write headnotes to stand alone without reference to the underlying opinion.6 Cf. Southco, Inc. v. Kanebridge Corp., 390 F.3d 276, 282 (3d Cir. 2004) (Alito, J.) (finding unoriginal product serial numbers in which “each number is rigidly dictated by . . . mechanical application of the [numbering] system, not creative thought” (citation modi- fied)). So the Second Circuit offers ROSS no refuge. B. Fair Use We turn next to ROSS’s purported fair-use defense. “[T]he fair use of a copyrighted work . . . is not an infringement of copyright.” 17 U.S.C. § 107. Fair use is an equitable doctrine that “permits courts to avoid rigid application of the copyright statute when, on occasion, it would stifle the very creativity 6 See Lexis Br. at 12 (explaining that Lexis, a Westlaw com- petitor, employs a different creative process as it generates dif- ferent headnotes for the same opinions). 13 which that law is designed to foster.” Andy Warhol, 598 U.S. at 527 (quoting Steward v. Abend, 495 U.S. 207, 236 (1990)). “Fair use is a mixed question of law and fact.” Harper & Row, 471 U.S. at 560. But where there are undisputed facts “suffi- cient to evaluate each of the statutory factors,” the Court “may conclude as a matter of law that the challenged use does not qualify as a fair use.” Id. (citation modified). In determining whether a use is fair, we must consider the following statutory factors: (1) the purpose and character of the use, including whether such use is of a commercial nature or is for non- profit educational purposes; (2) the nature of the copy- righted work; (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and (4) the effect of the use upon the potential market for or value of the copyrighted work. 17 U.S.C. § 107. As the proponent of this affirmative defense, ROSS bears the burden of demonstrating its copying was jus- tified. See Andy Warhol, 598 U.S. at 547 n.21; UpCodes, 172 F.4th at 261. We address each factor in turn. 1. Purpose and Character of the Use The first factor examines the “purpose and character” of ROSS’s use. 17 U.S.C. § 107(1); Andy Warhol, 598 U.S. at 14 532–33. To assess this factor, we first consider whether “a use is commercial as opposed to nonprofit” because a commercial use weighs against a finding of fair use. Andy Warhol, 598 U.S. at 531; see also UpCodes, 172 F.4th at 265 (“If a new work is used commercially rather than for a nonprofit purpose, its use will less likely qualify as fair.”). Next, we consider how transformative the copier’s use is—that is, “whether the new work merely supersedes the objects of the original crea- tion . . . or instead adds something new, with a further purpose or different character.” Andy Warhol, 598 U.S. at 528 (citation modified) (quoting Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 579 (1994)). The “lodestar” of this inquiry is whether the copier maintains a “distinctive purpose” for its use. 4 Nimmer on Copyright § 13F.10[G][2][a][i] (2026); see also Andy Warhol, 598 U.S. at 531 (“[A] use that has a distinct pur- pose is justified because it furthers the goal of copyright, namely, to promote the progress of science and the arts, with- out diminishing the incentive to create.”). And, because un- covering the differences between an originator’s and a copier’s purposes is “a matter of degree,” we ask “whether and to what extent the use at issue has a purpose or character different from the original.” Andy Warhol, 598 U.S. at 528–29 (citation mod- ified). “The larger the difference,” the Supreme Court has said, “the more likely the first factor weighs in favor of fair use.” Id. at 529. In sum, we must balance “the degree of difference [in pur- pose] . . . against the commercial nature of the use.” Id. at 532. And if “an original work and a secondary use share the same 15 or highly similar purposes, and the secondary use is of a com- mercial nature, the first factor is likely to weigh against fair use, absent some other justification for copying.” Id. at 532– 33. We start with whether ROSS’s use of Thomson Reuters’s headnotes is commercial. ROSS aimed to create a legal-re- search platform charging comparable prices to Westlaw to compete with Westlaw for customers. So ROSS’s use is com- mercial. But because commercialism is not “dispositive” of fair use, id. at 531, we must ask whether ROSS’s use of Thomson Reu- ters’s headnotes is transformative—that is “whether and to what extent” ROSS’s use of the headnotes “has a purpose or character different from” Thomson Reuters’s use. Id. at 529 (emphasis in original) (quoting Campbell, 510 U.S. at 579). Thomson Reuters uses its headnotes to help legal researchers find and understand judicial opinions containing points of law related to their searches. ROSS uses Thomson Reuters’s head- notes to create a legal-research platform to help users find ju- dicial opinion passages responsive to their searches. True, ROSS did not use the headnotes in precisely the same fashion as Thomson Reuters. ROSS took an intermediate step of using the headnotes to train an AI program, which arguably presents a slight degree of difference in use. But the undisputed evi- dence demonstrates that ROSS used the headnotes to train an AI program for the benefit of its legal-research platform. So both Thomson Reuters and ROSS use the headnotes to create 16 and optimize a legal-research platform that helps users find re- sponsive legal material. Thus, ROSS’s use of the headnotes shares the same ultimate purpose as Thomson Reuters’s use, making ROSS’s use minimally transformative, at best.7 7 We note that the United States Department of Justice (“DOJ”) filed a statement of interest relating to generative AI in a pend- ing case in the Southern District of New York. Statement of Interest of the United States, In re: OpenAI, Inc. Copyright In- fringement Litig., No. 1:25-md-3143 (S.D.N.Y. Sept. 1, 2026) (hereinafter “DOJ Statement __”). The concerns raised in that separate case do not apply here. The DOJ relied on Bartz v. Anthropic PBC, 787 F. Supp. 3d 1007, 1014 (N.D. Cal. 2025) to contend that training an AI large language model, which can “generate original responses,” is a transformative use. DOJ Statement at 10; see also Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026, 1034 (N.D. Cal. 2025) (exploring whether generative AI programs transformatively use training materi- als). Unlike the AI models in Bartz and In re: OpenAI, ROSS’s AI platform cannot generate original expression, and the evi- dence here supports the opposite conclusion about transforma- tiveness. The DOJ also argued that the large language model training in OpenAI did not result in “substitutive competition.” DOJ Statement at 13 (citation modified). Here, ROSS trained its AI for the purpose of creating a commercial substitute for Westlaw. In any event, the DOJ’s filing in In re: OpenAI demonstrates that the DOJ is tracking these issues and knows how to assert its interests, but the DOJ notably did not do so here. 17  Given that ROSS’s use was highly commercial and mini- mally transformative, we conclude that the first factor weighs against fair use. Andy Warhol, 598 U.S. at 537–38 (“Taken together, these two elements—that Goldsmith’s photograph and AWF’s 2016 licensing of Orange Prince share substan- tially the same purpose, and that AWF’s use of Goldsmith’s photo was of a commercial nature—counsel against fair use . . . .”); Fox News Network, LLC v. TVEyes, Inc., 883 F.3d 169, 178 (2d Cir. 2018) (reasoning that the “commercial nature of a secondary use weighs against a finding of fair use . . . espe- cially when . . . the transformative character of the secondary use is modest”). ROSS offers two arguments to support the opposite conclu- sion. Neither convinces. First, ROSS argues that Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015), dictates a different conclusion. There, the Second Circuit held that Google’s scanning of entire books had the “highly transforma- tive purpose” of creating a feature to search for and read small snippets of books that matched their search terms. Id. at 216– 17; see also Authors Guild, Inc. v. HathiTrust, 755 F.3d 87, 97 (2d Cir. 2014) (holding similar use was fair where there was no snippet feature and the use was noncommercial). Deciding a dispute that “test[ed] the boundaries of fair use,” the court held that the search feature served a completely different func- tion that was highly transformative of the original work. Au- thors Guild, 804 F.3d at 206, 216–17. The purpose of the orig- inal books was to be read while the purpose of the search 18 function was to enable users to find “books containing a term of interest to the searcher.” Id. at 216. No such transformative purpose is present here. Instead, ROSS’s platform uses Thomson Reuters’s headnotes to help users find judicial opinions related to their legal research in- quiries, something Thomson Reuters already does with its headnotes. Also favoring fair use in Authors Guild was that Google’s search feature may have driven users toward a book the users would then purchase from the original author. Id. at 217–18. Not so here. ROSS’s use does not lead a user to Westlaw; instead, as ROSS admits, it aims to replace Westlaw. Second, ROSS argues that three “intermediate copying” cases, one from the Supreme Court and two from the Ninth Circuit, dictate a different conclusion. Opening Br. 44. In a recent case, the Supreme Court held that Google could copy a portion of Oracle’s Java computer programming language to enable users to create applications using Java code on Android phones. Google LLC v. Oracle Am., Inc., 593 U.S. 1, 6–7 (2021). The Ninth Circuit concluded similarly in earlier cases. It held that a video game developer could copy Sega’s com- puter program code to create video games that could be played on Sega’s gaming system. Sega Enters. Ltd. v. Accolade, Inc., 977 F.2d 1510, 1514 (9th Cir. 1992). Another panel similarly held that Connectix could copy Sony’s computer program code to create a product that made Sony video games playable on personal computers. Sony Comput. Ent., Inc. v. Connectix Corp., 203 F.3d 596, 599, 606–08 (9th Cir. 2000). 19  In each of these cases, computer code was copied to enable users to make their software operable with an existing com- puter system. These uses were transformative because, in each case, copying was necessary to access the unprotected func- tional aspects of computer code. See Oracle, 593 U.S. at 30– 31 (stating that Google copied Oracle API “only insofar as needed” and explaining that “shared interfaces are necessary for different programs to speak to each other”); Sega, 977 F.2d at 1514 (noting that “no other means of access to the unpro- tected elements exists”); Connectix, 203 F.3d at 599 (holding that copying was “necessary to permit Connectix to make its non-infringing Virtual Game Station function with PlayStation games”).8 Here, ROSS does not need to copy Thomson Reu- ters’s headnotes to access the underlying unprotected infor- mation. ROSS had access to the underlying judicial opinions and could freely copy them to make the memos needed to train its AI. It chose not to do so because copying the headnotes offered an “easy” way to create its training memos. App. 8 Also critical to the outcome in Oracle was that the at-issue computer code was “declaring code,” the use of which “is in- herently bound together with uncopyrightable ideas” and that the value of Java as a program “lies in its efforts to encourage programmers to learn and to use that system.” 593 U.S. at 28– 29. By contrast, Thomson Reuters’s headnotes do not derive value from others copying them. 20 8545–46. Unlike necessity, ease is not a justification for cop- ying. In sum, the first factor weighs against fair use.9 2. Nature of the Copyrighted Work The second factor evaluates the “nature” of the headnotes. 17 U.S.C. § 107(2). Under this factor we consider whether the work was published or unpublished when it was used and if the work, while creative enough for copyright protection, has fac- tual or functional elements. Harper & Row, 471 U.S. at 563– 64. More creative, less factual works are “closer to the core of intended copyright protection.” Campbell, 510 U.S. at 586. This factor will generally weigh in favor of fair use when a 9 Courts have at times considered whether copying was done in good faith when evaluating the first factor. See Oracle, 593 U.S. at 32–33 (explaining the good faith consideration but ex- pressing skepticism about whether good faith still plays any role in the fair use analysis). Here, the undisputed evidence reflects that ROSS at times acted in bad faith. See, e.g., App. 4969–76, 6443, 6449, 6454, 6495 (ROSS attempting to access Westlaw with law-firm investor credentials despite notice that doing so was prohibited by Westlaw’s terms of service); App. 6491–93 (stating ROSS employee inquired about a Westlaw account under the guise of a solo practitioner); App. 6956 (stat- ing ROSS employee used student account to access Westlaw, hiding his employment by a competitor). So to the extent good faith remains relevant, that consideration weighs against fair use. 21 work is primarily factual or functional. Oracle, 593 U.S. at 20, 28–29. But courts have noted that “[t]he second factor has rarely played a significant role in the” fair use determination. Authors Guild, 804 F.3d at 220. Here, Thomson Reuters pub- lished the headnotes before ROSS used them. And the head- notes are more factual than fictional because they must accu- rately convey the law and describe uncopyrightable judicial opinions. This factor thus favors fair use. 3. Amount and Substantiality of the Portion Used For the third factor, we evaluate whether “the amount and substantiality of the portion used in relation to the copyrighted work as a whole” is “reasonable in relation to the purpose of the copying.” Campbell, 510 U.S. at 586 (quoting 17 U.S.C. § 107(3)). This factor requires us to assess, quantitatively, how much of the copyrighted work was used and, qualitatively, whether the use took “the heart” of the work. Harper & Row, 471 U.S. at 565. In evaluating the third factor, we “recognize that the extent of permissible copying varies with the purpose and character of the use.” Campbell, 510 U.S. at 586–87. As to the amount and substantiality of the headnotes used, ROSS copied the entire text of the 25,000 Westlaw-written headnotes into its memos and used those to train its AI pro- gram. ROSS’s purpose in copying the headnotes was highly similar to the headnotes’ original purpose, and ROSS aimed to make a commercial substitute for Westlaw using Westlaw’s work. See supra Section II.B.1. ROSS thus lacked a 22 transformative purpose and character that would make exten- sive copying reasonable. See Authors Guild, 804 F.3d at 221 (explaining that a transformative purpose allows more exten- sive copying than copying done to offer a competing substitute for the original). ROSS responds that it took an insubstantial portion of Thomson Reuters’s work—only 0.08% of Thomson Reuters’s 28 million headnotes. But copying even an “insubstantial por- tion” of an original work is not fair if, qualitatively, an im- portant part of the work is copied. Harper & Row, 471 U.S. at 564–65. “The crux of the inquiry is whether ‘no more was taken than necessary’” to achieve the copier’s purpose. Ha- thiTrust, 755 F.3d at 98 (quoting Campbell, 510 U.S. at 589). The purpose behind ROSS’s copying was minimally trans- formative, at best. Copying the headnotes was also not neces- sary to train ROSS’s AI, as the underlying judicial opinions were freely available. So, ROSS took more than necessary when it copied these headnotes without a justifying transform- ative purpose.10 Thus, the third factor weighs against fair use. 10 Additionally, we affirm the District Court’s holding that each individual headnote is a copyrightable work. See supra Sec- tion II.A; cf. Oracle, 593 U.S. at 33 (recognizing that taking one sentence from a book may seem insubstantial, unless that sentence itself constituted the entirety of a short story). So for each headnote taken, ROSS copied an entire work. Such ex- tensive copying was not reasonable because it was not sup- ported by a sufficiently transformative purpose. 23  4. Market Effect of the Use The fourth factor concerns “the effect of the use upon the potential market for or value of the copyrighted work.” 17 U.S.C. § 107(4). The “potential market” includes both the market for the original work and any market for derivative uses of the original work. Campbell, 510 U.S. at 590 (quoting 17 U.S.C. 107(4)). In this context, we will also consider any “public benefits the copying will likely produce.” Oracle, 593 U.S. at 35. So to resolve this appeal, we consider harm to the original market, harm to the value of the copyrighted work, harm to the potential derivative market, and the alleged public benefits of the copying. We begin with the original market. The parties dispute which is the relevant original market. ROSS argues that the relevant original market is the market for headnotes as a standalone product, which ROSS claims does not exist. Even assuming this is the relevant market, ROSS’s argument ignores the effect of ROSS’s use on the value of Thomson Reuters’s headnotes. Our Court rejected a similar argument in Video Pipeline, Inc. v. Buena Vista Home Entertainment, Inc., 342 F.3d 191, 202 (3d Cir. 2003), abrogated in part on other grounds by TD Bank N.A., 928 F.3d at 278. There, a website that published trailers and clips of Disney movies argued that no market existed for movie trailers because no one “ever paid or will ever pay any money merely to see trailers.” Id. at 202. We rejected that argument, stating that “[t]he [fair use] statute directs us to consider the effect of the use upon the . . . value 24 of the copyrighted work, not only the effect upon the market, however narrowly that term is defined.” Id. (emphasis in orig- inal) (citation modified). Video Pipeline’s publishing of the trailers damaged the trailers’ value in drawing users to Dis- ney’s website, where Disney could induce purchases or cross- market other products; it made no difference that no standalone trailer market existed. Id. The same is true here. While there may be no standalone market for headnotes, Thomson Reuters touts the headnotes as a feature to attract users to buy a Westlaw subscription. By copying the headnotes and using them to build its own competing legal-research platform, ROSS appropriated the headnotes’ value for finding and ana- lyzing judicial opinions and diminished their value as a draw for users to Thomson Reuters’s legal-research platform. So, Video Pipeline supports the conclusion that ROSS’s copying harmed the value of Thomson Reuters’s headnotes. Thomson Reuters argues that the legal-research platform market is the relevant original market. ROSS offers no con- vincing reason to ignore this market, so we consider it. The undisputed evidence here reveals that ROSS copied the head- notes to create a competing legal-research platform in the same market as Westlaw. “[U]nrestricted and widespread” copying of Thomson Reuters’s headnotes to develop substitute legal- research tools, like ROSS did, would harm Westlaw’s position in that original market. Andy Warhol, 598 U.S. at 524 n.3 (quoting Campbell, 510 U.S. at 590). ROSS offers no evidence to rebut the alleged harm in this market. 25  We turn now to potential derivative markets for Thomson Reuters’s headnotes. The parties dispute the relevant potential derivative market. ROSS argues it did not harm any potential derivative market because no derivative market for headnotes exists. Thomson Reuters argues that ROSS harmed its position in the potential derivative market for licensing headnotes as AI training data. “The market for potential derivative uses in- cludes only those that creators of original works would in gen- eral develop or license others to develop.” Campbell, 510 U.S. at 592. But a potential derivative market is not illusory just because an author “has evidenced little if any interest in ex- ploiting this market for derivative works.” Castle Rock Ent., Inc. v. Carol Publ’g Grp., Inc., 150 F.3d 132, 145–46 (2d Cir. 1998). Here, the evidence shows that the market for licensing headnotes as text to train AI is rapidly developing. Addition- ally, Thomson Reuters presented evidence that it is using its headnotes as training data for its own AI search products. ROSS offers no evidence to disprove this. That Thomson Reu- ters did not license its headnotes to others does not disprove that a market exists to do so. Additionally, ROSS’s copying harmed Thomson Reuters in this market. By copying the head- notes for use as training data without Thomson Reuters’s au- thorization, ROSS usurped Thomson Reuters’s opportunity to enter that derivative market and license its headnotes for that purpose. The market harm caused by ROSS’s use weighs against fair use. 26  Finally, we consider whether the public benefits are “re- lated to copyright’s concern for the creative production of new expression” and how “comparatively important, or unim- portant” the public benefits are “when compared with dollar amounts likely lost” due to copying. Oracle, 593 U.S. at 35– 36. ROSS claims its copying will benefit the public by increas- ing access to the law. But the underlying judicial opinions are freely available, and ROSS’s service was offered at prices comparable to Westlaw. So it is unclear how ROSS’s copying would greatly increase public access to the law. ROSS also argues the District Court’s decision will halt AI development. But ROSS presents no evidence to support that broad claim. Finally, ROSS argues that AI is critical for national security purposes so copyright should not impede ROSS’s AI develop- ment. ROSS again presents no evidence connecting its legal- research platform to national security. Despite doing so in other circumstances,11 the DOJ has not raised national security concerns about the conduct in this case. While some AI tech- nology may implicate national security concerns, this does not give ROSS carte blanche to violate copyright law merely be- cause it incorporates AI. Thus, the fourth factor weighs against fair use. In conclusion, the second factor weighs slightly in favor of fair use, but the first, third, and fourth factors weigh against it. Evaluating these factors together, and in light of copyright’s purposes, we hold that ROSS’s use was not fair. See UpCodes, 11 See, e.g., DOJ Statement at 1. 27 172 F.4th at 261 (exploring and weighing all factors “together, in light of the purposes of copyright”) (quoting Murphy v. Mil- lennium Radio Grp. LLC, 650 F.3d 295, 306 (3d Cir. 2011)). ***** Because Thomson Reuters’s 2,243 headnotes are original enough for copyright protection and ROSS’s use of the head- notes was not fair, we will AFFIRM. 28 Counsel for Appellant Ji Won Oh Kufere Laing Mark S. Davies [Argued] Andrew M. Legolvan Yar R. Chaikovsky WHITE & CASE Kayvan M. Ghaffari Anne M. Voigts Ranjini Acharya PILLSBURY WINTHROP SHAW PITTMAN Anna Naydonov CLEARY GOTTLIEB STEEN & HAMILTON Counsel for Appellees Dale M. Cendali [Argued] Joshua L. Simmons Miranda D. Means KIRKLAND & ELLIS Counsel for Amici Appellant Heather J. Meeker TECHNOLOGY LAW PARTNERS 29 Victoria J. Noble ELECTRONIC FRONTIER FOUNDATION Rebecca L. Tushnet HARVARD LAW SCHOOL Karen E. Keller SHAW KELLER Tim R. Hwang HWANG INDUSTRIES Abraham J. Kang LAW OFFICE OF ABRAHAM KANG Michael P. Abate KAPLAN JOHNSON ABATE & BIRD Michael T. Hensley CARLTON FIELDS Michael S. Kwun KWUN BHANASALI LAZARUS August Gebhard-Koenigstein Phillip R. Mallone STANFORD LAW SCHOOL INTELLECTUAL PROPERTY AND INNOVATION CLINIC, MILLS LEGAL CLINIC 30 Counsel for Amici Appellees Lawrence J. Spiwak PHOENIX CENTER FOR ADVANCED LEGAL & ECONOMIC PUBLIC POLICY STUDIES Lucy Grace D. Noyola ASSOCIATION OF AMERICAN PUBLISHERS J. Kevin Fee Jane Wise Stanley Panikowski DLA PIPER Jacqueline Charlesworth BAYLA LAW GROUP Nicholas M. Medellin EISNER Regan A. Smith NEWS MEDIA ALLIANCE Nancy E. Wolff COWAN DEBAETS ABRAHAMS & SHEPPARD Michael D. Hobbs Austin D. Padgett Brooke R. Watson 31 TROUTMAN PEPPER LOCKE Irina Tarsis CENTER FOR ART LAW Tod H. Cohen AI COALITION FOR DATA INTEGRITY Jonathan Iwry Jonathan Z. King COWAN LIEBOWITZ & LATMAN Felicity Kohn PRYOR CASHMAN Adam G. Unikowsky JENNER & BLOCK 32