Asutay Duhan Meydan / Attorney at law
The decision in Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc. is significant for reasons that go beyond artificial intelligence.
On 29 September 2026, the U.S. Court of Appeals for the Third Circuit affirmed the District Court’s grant of partial summary judgment in favor of Thomson Reuters. The court held that 2,243 Westlaw headnotes were sufficiently original to receive copyright protection and that ROSS Intelligence’s use of those headnotes in developing an AI-powered legal research platform did not constitute fair use. Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., No. 25-2153 (3d Cir. Sept. 29, 2026).
The case is frequently described as an “AI copyright case.” That description is correct, but incomplete. ROSS was not operating a generative AI model comparable to a large language model. Its system did not generate new legal prose. It returned relevant passages from existing judicial opinions. The Third Circuit itself stressed this distinction and expressly declined to treat the case as resolving the broader copyright questions now arising around generative AI training.
What makes the decision particularly interesting from the perspective of Turkish law is not simply the outcome. It is the route by which the U.S. court reached it.
Two concepts stand out. The first is substantive: fair use and, within that analysis, transformative use. The second is procedural: partial summary judgment, through which the court was able to resolve central questions of copyright liability without waiting for every remaining issue in the litigation to be tried.
Neither concept has a direct equivalent in Turkish law.
That does not mean a Turkish court confronted with the same facts would necessarily reach a different result. It means that it would ask different questions, under a different statutory structure, and reach its judgment through a different procedural path.
I. Fair Use and Transformative Use: What Would a Turkish IP Court Ask Instead?
The Third Circuit began with a familiar four-factor inquiry under §107 of the U.S. Copyright Act. It considered the purpose and character of ROSS’s use, the nature of the copyrighted works, the amount and substantiality used, and the effect of that use upon the actual and potential market for the works.
The first factor became particularly important because of transformative use.
The court asked whether ROSS had given the Westlaw headnotes a genuinely different purpose or character. Thomson Reuters used those headnotes to assist legal researchers in identifying and understanding relevant judicial decisions. ROSS used them as part of the process of developing another legal research platform designed to identify responsive judicial passages.
There was a technical difference: ROSS inserted an AI-training stage between the headnotes and the product ultimately offered to users. But the court considered that difference insufficient. Both uses ultimately served substantially the same legal-research function. ROSS was also developing a commercial product intended to compete directly with Westlaw. The court therefore characterised the use as, at best, only minimally transformative.
The remaining factors reinforced that conclusion. Although the factual nature of the headnotes slightly favored ROSS, the Third Circuit found that the amount used, the absence of necessity for copying the headnotes and, particularly, the market effect weighed against fair use. ROSS could have used the underlying judicial opinions, which were freely available. Instead, it used the editorial work Thomson Reuters had already performed. The court also found an emerging market for licensing headnotes as AI training material and held that unauthorized copying could interfere with that potential market.
The final balance was therefore clear: the first, third and fourth factors weighed against ROSS; only the second weighed slightly in its favor. The use was not fair.
A Turkish court would not perform this four-factor analysis.
There Is No General Fair Use Doctrine Under the FSEK
The Turkish Act on Intellectual and Artistic Works No. 5846 (“FSEK”) does not contain an open-ended provision comparable to §107 of the U.S. Copyright Act.
The structure is fundamentally different. A Turkish court would first determine whether the material in question qualifies as a protected work; then whether an exclusive economic right has been exercised without authorization; and finally whether the particular use falls within one of the statutory limitations or exceptions.
The inquiry would therefore begin with the Westlaw headnotes themselves.
Under Article 1/B of the FSEK, a work must carry the individuality of its author — sahibinin hususiyetini taşıması — and fall within one of the statutory categories of protected works. Article 2 includes works expressed through language and writing within scientific and literary works.
An important distinction would immediately arise.
The underlying judicial decisions themselves would not present the same copyright issue. Article 31 of the FSEK expressly provides that officially published or announced judicial decisions may be reproduced, distributed, adapted and otherwise used freely.
But a privately prepared editorial headnote is not the judicial decision itself.
Westlaw’s editors selected which legal propositions were sufficiently important to deserve a headnote, decided how much factual context was necessary and drafted a concise expression capable of standing independently from the full judgment. The Third Circuit found precisely these editorial choices sufficient to satisfy the U.S. originality threshold.
A Turkish IP court would use the language of hususiyet, not “creative spark.” Yet the underlying inquiry would be recognisable. If the particular wording, selection and condensation of the judicial material reflected sufficiently individual intellectual choices by the editor, there would be a strong basis for treating those headnotes as protected scientific or literary works under Articles 1/B and 2.
Article 6 could also become relevant. Turkish law recognises adaptations created by making use of another work, including explanations, commentaries and abridgements. Such an adaptation may itself receive protection if it carries the adapter’s individuality.
The precise classification of a Westlaw headnote — independent literary work or protected adaptation — would depend on the individual text. But the important point is that Article 31’s freedom to use the judicial decision would not automatically extend to the private editor’s own expression.
AI Training Would Not Make the Reproduction Disappear
The next question would be considerably more straightforward.
Article 22 of the FSEK grants the author the exclusive right to reproduce a work, in whole or in part, by any means or method, directly or indirectly, temporarily or permanently.
The ROSS record is unusual because the evidentiary problem that complicates many modern generative-AI cases was largely absent. This was not merely an allegation that a protected work had somehow influenced model parameters.
ROSS had engaged LegalEase to prepare approximately 25,000 training memoranda. Westlaw headnotes were used in producing those memoranda, and the District Court identified 2,243 headnotes for which the similarity was such that no reasonable juror could find that copying had not occurred.
In a Turkish proceeding, that factual record would strongly direct the analysis toward Article 22.
The fact that the headnotes were used at an intermediate stage of the AI-development process rather than displayed verbatim to the final user would not, by itself, eliminate the reproduction. If the protected text was copied into training memoranda or another storage medium, the relevant act of reproduction may already have occurred.
This is one of the points at which the American and Turkish analyses begin to diverge sharply.
“Transformative Use” Is Not the Same Thing as “İşleme”
At first glance, it may be tempting to translate the American concept of transformative use into the Turkish copyright concept of işleme.
That would be legally misleading.
Under U.S. law, transformation may support the lawfulness of an otherwise unauthorized use. The more genuinely different the secondary purpose or character, the stronger the argument under the first fair-use factor.
Under Turkish law, transformation does not operate as a general defence.
Article 21 provides that the right to exploit a work by adapting it belongs exclusively to the author. Article 6 defines an adaptation as a work produced by making use of another work, which is not independent of the original and which carries the adapter’s individuality.
The two concepts therefore operate in very different directions.
In American law, a sufficiently transformative use may assist the defendant.
In Turkish law, if the transformation remains legally dependent upon the original work, it may bring the defendant within another exclusive economic right.
This does not mean that every technical transformation of data is an “işlenme” under Article 6. Nor does it mean that every trained AI model is necessarily an adaptation of its training materials. Those questions require a separate examination of whether protected expression remains identifiable and whether the later product is legally dependent upon the original work.
But it does mean that a defendant before a Turkish court could not simply argue:
We changed the work considerably, therefore our use was transformative and lawful.
The relevant question would instead be:
What protected expression was reproduced or adapted, and where is the statutory authorization for that use?
That is a very different starting point.
Could ROSS Rely on a FSEK Exception?
On the facts considered by the Third Circuit, the obvious exceptions appear difficult to invoke.
Article 35 permits quotation in specifically defined circumstances. It allows, among other things, certain portions of a disclosed work to be incorporated into an independent scientific or literary work and requires the quotation to be identifiable and, for scientific works, the source and author to be indicated.
The systematic copying of complete headnotes for the purpose of constructing training material for a commercial competing product is difficult to characterise as quotation in this sense.
Article 38 would be still less persuasive. It permits reproduction for personal use provided that it is carried out without a profit-making purpose and does not unjustifiably prejudice the legitimate interests of the right holder or conflict with the normal exploitation of the work. ROSS was developing a commercial legal-research platform intended to compete with Westlaw.
Most importantly, the FSEK presently contains no general equivalent of American fair use under which a court could simply balance the commercial purpose, transformation, amount copied and market effect and declare an otherwise unauthorized reproduction “fair.”
That structural difference could make the Turkish analysis more direct than the U.S. analysis.
What Might the Reasoning of a Turkish IP Court Look Like?
Assuming that Turkish law governed the dispute and the evidentiary record were materially the same, the reasoning of an Intellectual and Industrial Property Rights Court could plausibly take the following form:
The judicial decisions underlying the disputed headnotes may be freely used pursuant to Article 31 of the FSEK. The dispute, however, does not concern the reproduction of those judicial decisions as such. It concerns editorial texts prepared through the selection, condensation and individual expression of legal propositions contained in those decisions.
To the extent that the disputed headnotes carry the individuality of their authors, they constitute protected works within the meaning of Articles 1/B and 2 of the FSEK.
The defendant caused those protected texts to be reproduced in training memoranda used for the development of its commercial legal-research system. Pursuant to Article 22, reproduction includes the direct or indirect, temporary or permanent reproduction of a work in whole or in part by any means or method.
The fact that the copies were used during the training process and were not ultimately displayed as headnotes to the end user does not retrospectively remove the act of reproduction.
Nor does Turkish copyright law contain a general rule under which an unauthorized reproduction becomes lawful because the subsequent use is technologically transformative. The concept of “transformative use” developed within U.S. fair-use jurisprudence has no direct equivalent in the FSEK. To the contrary, where a secondary work remains dependent upon the original, adaptation may itself fall within the exclusive right regulated by Article 21.
Since the use at issue is commercial and was undertaken for the development of a competing legal-research service, Article 38 concerning personal use is inapplicable. The manner and extent of copying likewise do not satisfy the conditions of Article 35 concerning quotation.
Accordingly, in the absence of authorization from the right holder or another applicable statutory limitation, the reproduction constitutes an infringement of the relevant economic right.
That reasoning would not require the Turkish judge to decide every theoretical question about AI training. In particular, once unlawful reproduction of the actual headnotes had been established, the court might have no need to decide whether the trained model itself constituted an adaptation under Article 21.
That is another reason the ROSS case should not be read as a general answer to copyright and generative AI.
The decisive evidence was considerably narrower.
II. The Other Major Difference: Turkey Has No Rule 56-Style Partial Summary Judgment
The second comparative point is procedural.
Federal Rule of Civil Procedure 56 expressly allows a party to seek summary judgment not only on an entire claim or defence but also on part of a claim or defence. The test is whether there is any genuine dispute as to a material fact and whether the moving party is entitled to judgment as a matter of law.
That mechanism mattered greatly in ROSS.
The District Court did not need to wait until every remaining factual and remedial question in the litigation had been tried before deciding that 2,243 headnotes were protected, had been copied and that ROSS’s fair-use defence failed.
It entered partial summary judgment on those matters.
The parties then requested interlocutory appellate review of two questions: the originality of the headnotes and whether ROSS’s use constituted fair use. The Third Circuit accepted jurisdiction under 28 U.S.C. §1292(b) and affirmed.
It is worth stressing that a partial summary judgment is not automatically an independently appealable final judgment. Rule 54(b) separately governs entry of final judgment as to fewer than all claims or parties, and Rule 56 itself contemplates interlocutory issue determination. The ROSS appeal reached the Third Circuit through a specifically certified interlocutory appeal.
There is no direct procedural equivalent under the Turkish Code of Civil Procedure.
Article 294 of the HMK adopts the ordinary model under which the court terminates the action through a final procedural or substantive decision, and the judgment on the merits is rendered at the end of the proceedings. Article 297 further requires the operative part of the judgment to address the claims presented and to state the rights and obligations of the parties clearly and without ambiguity.
There is therefore no general Turkish motion by which a party may say:
“There is no genuine dispute of material fact concerning copyright liability. Enter judgment on liability now and leave damages and the remaining issues for later.”
A Turkish court may of course issue interlocutory orders throughout the proceedings. It may decide procedural questions, appoint experts, determine which evidence will be collected and separate joined actions where the requirements of Article 167 are satisfied. A claimant may also bring a partial action under Article 109 where the subject matter is divisible.
None of these institutions is equivalent to Rule 56.
The distinction is particularly visible in arbitration. Article 436(2) of the HMK expressly provides that, unless otherwise agreed, arbitrators may issue partial awards. The legislative materials expressly contemplate partial decisions on issues such as jurisdiction, liability and the amount of damages.
No comparable express mechanism exists for ordinary state-court litigation.
There is, it should be added, a genuine doctrinal discussion in Turkish civil procedure as to whether a state court may issue a kısmi hüküm in certain circumstances, particularly where several independent claims are joined and one has become ripe for decision before the others. Turkish scholarship therefore does not justify the absolute proposition that the concept of a partial judgment is wholly unknown.
What is absent is the American Rule 56 mechanism: a codified procedure enabling the court to dispose of part of a claim or defence because no genuine dispute of material fact exists.
That distinction matters.
If the ROSS litigation were conducted before a Turkish IP court as a single infringement and damages action, the ordinary expectation would be that the court would complete the evidentiary phase, determine whether the headnotes were protected, establish whether copying had occurred, assess the available defences and exceptions, obtain the evidence necessary for the requested remedies and then give its substantive judgment.
The liability question would ordinarily become part of that final judgment rather than being converted, mid-proceedings, into a Rule 56-style binding merits determination while the remainder of the same claim continued.
The resulting judgment would then be subject to the ordinary Turkish appellate structure. Final judgments of courts of first instance are, subject to the statutory conditions of appeal, reviewed by the Regional Courts of Appeal through istinaf under Article 341 HMK. Decisions of the Regional Courts of Appeal that are eligible for further review may then be brought before the Court of Cassation under Article 361.
In other words, the Turkish system does not eliminate appellate review. It places the principal merits determination at a different procedural point.
The Same Dispute, Two Different Legal Questions
The most instructive part of Thomson Reuters v. ROSS is therefore not that an American court “rejected AI training.”
It did not.
The Third Circuit expressly separated ROSS from the generative-AI cases now being litigated in the United States. ROSS did not generate original expression; it used protected editorial material to help construct a commercial substitute operating in substantially the same legal-research market.
What the decision demonstrates is how differently the same conduct can be framed across legal systems.
An American court asks whether the unauthorized use can nevertheless be justified as fair use, with transformative purpose playing an important role in that balancing exercise.
A Turkish court would begin elsewhere.
It would ask whether the headnotes constitute protected works, whether protected expression was reproduced or adapted, and whether the defendant can identify a specific statutory limitation allowing that use.
That difference also exposes a frequent translation error in comparative AI copyright discussions. Transformative use is not simply the American name for the Turkish concept of “işleme.” One may function as an argument supporting permissible use; the other describes conduct falling within an author’s exclusive economic rights.
The procedural comparison is equally revealing.
The American court was able to determine important aspects of liability before the entire litigation had concluded. Turkish state-court procedure contains no comparable Rule 56 architecture. The merits of an infringement action will ordinarily be determined through the final judgment after the court has completed its examination of the case, with istinaf and, where available, cassation review remaining open thereafter.
The likely substantive destination in a Turkish version of ROSS might therefore be similar.
The road to that destination would not be.
Sources and Legal Basis
The analysis in this article is based primarily on the following judicial decisions, statutory provisions and procedural rules.
United States
Thomson Reuters Enterprise Centre GmbH; West Publishing Corp. v. ROSS Intelligence Inc., No. 25-2153, U.S. Court of Appeals for the Third Circuit, Opinion filed 29 September 2026. The Third Circuit affirmed the District Court’s conclusions concerning the copyrightability of the relevant Westlaw headnotes and rejected ROSS Intelligence’s fair-use defence. The opinion is the principal source for the discussion of transformative use, market substitution, AI training and the procedural history of the dispute.
17 U.S.C. § 107 — Limitations on Exclusive Rights: Fair Use. The statutory basis of the U.S. fair-use doctrine. Section 107 requires consideration of the purpose and character of the use, the nature of the copyrighted work, the amount and substantiality used, and the effect of the use upon the potential market for or value of the copyrighted work.
Federal Rule of Civil Procedure 56 — Summary Judgment. Rule 56 expressly permits a party to seek summary judgment as to an entire claim or defence, or only part of a claim or defence, where there is no genuine dispute as to any material fact and the moving party is entitled to judgment as a matter of law. This is the procedural basis for the concept commonly referred to as partial summary judgment.
Federal Rule of Civil Procedure 54(b) — Judgment on Multiple Claims or Involving Multiple Parties. Rule 54(b) regulates when a court may direct entry of a final judgment as to fewer than all claims or parties. It is important to distinguish such a final judgment from an interlocutory partial summary judgment under Rule 56.
28 U.S.C. § 1292(b) — Interlocutory Appeals. This provision allows, under defined conditions and with the permission of the Court of Appeals, immediate appellate review of an otherwise non-final order involving a controlling question of law. It provides the relevant statutory framework for understanding how legal questions may reach an appellate court before the entire litigation has been finally resolved.
Turkey
Law No. 5846 on Intellectual and Artistic Works — Fikir ve Sanat Eserleri Kanunu (“FSEK”).
The Turkish-law analysis principally relies upon:
Article 1/B — statutory definitions, including the requirement that a protected work carry the individuality of its author (sahibinin hususiyetini taşıması);
Article 2 — scientific and literary works expressed through language and writing;
Article 6 — adaptations (işlenmeler) and the requirement that an adaptation retain a legally relevant relationship with the original work;
Article 21 — the author’s exclusive right of adaptation (işleme hakkı);
Article 22 — the author’s exclusive right of reproduction, including direct or indirect and temporary or permanent reproduction by any means or method;
Article 31 — freedom to reproduce, distribute, adapt or otherwise use officially published or announced legislation, judicial decisions and other official texts;
Article 35 — freedom of quotation (iktibas serbestisi) subject to the statutory conditions;
Article 38 — reproduction for personal use, subject to the absence of a profit-making purpose and the requirement that the use not unjustifiably prejudice the legitimate interests of the right holder or conflict with the normal exploitation of the work.
There is no general statutory fair-use doctrine in the FSEK corresponding to 17 U.S.C. § 107. Accordingly, references in this article to the likely treatment of ROSS under Turkish copyright law are a comparative legal analysis based on the structure of the FSEK and should not be understood as suggesting that U.S. “transformative use” and Turkish işleme are equivalent legal concepts.
Code of Civil Procedure No. 6100 — Hukuk Muhakemeleri Kanunu (“HMK”).
The procedural comparison principally relies upon:
Article 109 — partial actions (kısmi dava);
Article 167 — separation of joined actions;
Article 294 — the nature of judgments and the termination of proceedings through final procedural or substantive decisions;
Article 297 — the required content and operative part of a judgment;
Article 341 — judgments and decisions subject to review by the Regional Courts of Appeal (istinaf);
Article 361 — decisions of the Regional Courts of Appeal subject to review by the Court of Cassation (temyiz);
Article 436(2) — expressly allowing arbitral tribunals, unless otherwise agreed by the parties, to render partial awards.
The Turkish Code of Civil Procedure does not establish a general state-court mechanism equivalent to Federal Rule of Civil Procedure 56 under which a court may enter summary judgment on part of a claim or defence solely because no genuine dispute of material fact remains.
This should be distinguished from the concepts of a partial action, separation of proceedings, interlocutory procedural decisions and the doctrinal discussion concerning kısmi hüküm in Turkish civil procedure. It should also be distinguished from arbitration, for which Article 436(2) HMK expressly recognises the possibility of partial awards.
Accordingly, the procedural conclusions in this article concern the absence in Turkish state-court litigation of a Rule 56-style partial summary judgment mechanism, rather than an assertion that Turkish procedural law is entirely unfamiliar with decisions addressing only part of a broader dispute.