The Turkish counterparts of the pleaded claims and the likely judicial approach
Asutay Duhan Meydan
Attorney at Law | Meydan Law Office
8 October 2026
On 8 October 2026, USA TODAY Co., Inc. and affiliated newspaper companies filed a copyright action against seven OpenAI entities in the Southern District of New York. The complaint concerns content from nineteen publications and seeks more than USD 250 million, injunctive relief and destruction of models and training sets incorporating the plaintiffs’ content. These are pleaded allegations and requested remedies, not judicial findings. [1, paras 1–16, 187–209; Prayer for Relief]
For Turkey, the claim has recognisable legal counterparts. Unauthorised copying would principally engage the reproduction right under Article 22 of Law No. 5846 on Intellectual and Artistic Works (FSEK). Removal of rights information has a specific counterpart in Additional Article 4. Liability across the corporate group would require a separate basis for each defendant. Neither the American damages calculation nor the requested destruction of an entire model would transfer automatically into Turkish law.
The claims that would have to be translated
The complaint pleads three counts: direct copyright infringement, vicarious copyright infringement and removal of copyright management information under the Digital Millennium Copyright Act (DMCA). It alleges copying during data acquisition and training, retention of protected material in model parameters, and reproduction or repackaging in outputs. Its search examples involve retrieval augmented generation (RAG), through which external material is supplied to a model when answering a query. [1, paras 113–141, 187–209]
The direct infringement claim under 17 USC §§ 106 and 501 would principally correspond to FSEK Article 22 on reproduction, Article 21 on adaptation and Article 25 on communication to the public, according to the particular conduct proved.
The vicarious infringement claim would require examination of the enterprise liability contemplated by FSEK Article 66 and, for damages, the relevant provisions of the Turkish Code of Obligations (TBK), particularly Articles 49, 61 and, where applicable, 66.
The removal of copyright management information alleged under § 1202 would find its closest Turkish counterpart in FSEK Additional Article 4, supplemented where relevant by Articles 15, 16 and 67 concerning attribution, alteration and remedies for infringement of moral rights.
The territorial premise
This analysis assumes equivalent conduct for which protection is sought in Turkey. Article 23(1) of Law No. 5718 on Private International Law and International Civil Procedure (MÖHUK) applies the law of the country for which intellectual property protection is claimed. Turkish jurisdiction is a separate question under Article 40.
A service being accessible in Turkey does not, without further analysis, place every overseas training operation under FSEK. The location and character of the alleged acts, the rights invoked and the territorial scope of the relief must be identified. [4]
The analysis develops my earlier examination of The New York Times v OpenAI. Its distinction between data acquisition, training and user-facing outputs remains essential here: liability at one stage does not establish liability at every other stage. [10]
Protected journalism and the reproduction right
A Turkish court would first identify the protected works and the claimant’s entitlement to enforce the relevant rights. Under FSEK Articles 1/B and 2, an article must bear its author’s individuality and fall within a statutory category of works. The event reported, underlying facts and general information do not become the publisher’s exclusive property merely because its journalists discovered them. Original wording, selection and arrangement may nevertheless qualify for protection. [2]
The journalist is ordinarily the author under Article 8. Article 18 provides, subject to contrary agreement or the nature of the work, for employers to exercise economic rights in works created by employees in performing their duties.
Freelance contributions and acquired rights require separate examination. Article 52 requires transactions concerning economic rights to be in writing and to specify the rights individually. Publication on a newspaper website therefore cannot substitute for proof of entitlement. American registrations would be evidence, not a replacement for this inquiry. [2]
Article 22 covers complete or partial, direct or indirect, temporary or permanent reproduction by any method. Downloading protected articles, retaining dataset copies and reproducing them during processing or training can fall within that wording. The relevant question is whether protected expression was copied, not whether the developer ultimately offered readers a conventional article archive.
The complaint alleges more than 160,000 entries from the plaintiffs’ publications in WebText, including 83,266 from usatoday.com. Such allegations would require authentication and linkage to the works and operations actually challenged. [1, paras 100–109; 2]
A third-party dataset does not itself confer a licence. Conversely, familiarity with reported facts does not prove that a particular article was copied. Nor should a court assume that every model parameter is a copy of a protected work. That allegation requires technical evidence of what is encoded and whether protected expression can be reconstructed.
Dataset copies and model weights should not receive an identical legal characterisation merely because they belong to the same development process.
The defence would turn on specific exceptions
American fair use under 17 USC § 107 evaluates purpose, the nature of the work, the amount taken and market effects. FSEK contains no equivalent general balancing provision. Technological innovation or a transformative purpose would therefore not, by itself, create a Turkish exception authorising commercial training.
Article 38 permits non-profit personal reproduction, subject to the right holder’s legitimate interests and normal exploitation. A provider’s commercial training operation does not become personal use because individual consumers later use the product. [2, 3]
News-related exceptions require closer attention. Article 36 allows quotation of daily news and regulates reuse of articles on current social, political and economic matters. Within that provision, abbreviated press summaries remain permissible even where quotation rights have been reserved; the prescribed source and author information must be given.
Article 35 separately regulates quotation. Article 37 allows parts of works to be used in reporting current events within the informative purpose, without harming legitimate interests or normal exploitation. [2]
These provisions may justify particular outputs. They do not expressly authorise building a permanent commercial training corpus from an entire newspaper archive. Each claimed exception must be tested against its own wording, subject matter and conditions.
A robots.txt exclusion or website prohibition is relevant evidence of reservation and knowledge, but does not automatically extinguish a statutory exception or establish a binding contract.
An output must be examined against the original article
The output claim would principally engage Article 22 where protected passages are reproduced. Article 21 may apply to adaptations derived from a protected work; Article 25 may apply to the manner in which protected material is communicated or made available.
Neither rearrangement nor an extensive summary automatically establishes adaptation or infringement. The court must identify retained protected expression and examine any applicable exception. [2]
This distinction is supported by a relevant American ruling, independently of the present complaint. On 4 April 2025, Judge Sidney H. Stein dismissed the Center for Investigative Reporting’s claims concerning identified abridgments in the related OpenAI litigation. The particular summaries conveyed facts in different wording, arrangement, style and sentence structure and were not substantially similar to protected expression.
This was a ruling on specified claims at the pleading stage, not a general approval of AI summarisation, and it does not bind a Turkish court. [8, pp 41–42]
USA TODAY alleges that its examples follow the original articles’ organisation and provide substitutes for reading them. A Turkish court would need the full source articles, the complete prompts and outputs, and a comparison that distinguishes factual sequence from original expression.
Lost visits or subscriptions could be relevant to injury, but an economically useful summary is not necessarily an infringing summary. [1, paras 117–141]
Rights information has a direct statutory counterpart
Additional Article 4 FSEK prohibits unauthorised removal or alteration of information identifying the work, its author, a rights holder, or the terms and conditions of use, including associated numbers and codes appearing on copies or during communication to the public.
It also prohibits specified dissemination of originals or copies from which that information has been removed or altered. This is the closest statutory counterpart to USA TODAY’s § 1202 claim. [2, 3]
The complaint alleges that text-extraction tools deliberately removed bylines, copyright notices and related identifiers during dataset preparation and that outputs omitted such information.
Turkish proof would have to connect an identified protected work and its rights information to an unauthorised removal, alteration or prohibited dissemination. A generic allegation that models do not ordinarily reproduce metadata is insufficient to establish that entire chain.
The Turkish provision does not reproduce the American statute’s formulation requiring knowledge of removal and knowledge, or reasonable grounds to know, of its connection with infringement. [1, paras 201–209; 2, 3]
Article 15 on identification of the author and Article 16 on unauthorised alteration may additionally apply, with Article 67 providing relevant cessation and correction remedies.
Those moral rights must be distinguished from the publisher’s economic rights. Their enforcement depends on the claimant’s legal authority; they do not become the corporate publisher’s rights merely because it owns or exercises economic rights. [2]
Group membership is not a basis of liability by itself
Count II alleges that certain OpenAI entities directed, controlled and profited from infringement by operating entities. Turkish law would require the acts and legal responsibility of each defendant to be established.
FSEK Article 66 allows cessation proceedings against the enterprise owner for infringements committed by its representatives or employees in performing their duties, without requiring fault for that relief. [1, paras 198–200; 2]
For damages, TBK Article 49 supplies fault-based tort liability, Article 61 addresses persons jointly causing or otherwise liable for the same damage, and Article 66 may apply to an employer subject to its statutory conditions and defences.
Corporate ownership, receipt of profits and a capacity to influence a subsidiary do not alone establish joint and several liability. Instructions to acquire protected data, actual participation and responsibility for the resulting harm would matter. [5]
Additional Turkish claims and the evidence needed
A Turkish publisher could also consider rights in its archive. Article 6 FSEK can protect an original selection and arrangement of database contents.
Additional Article 8 separately protects a database producer making substantial qualitative or quantitative investment in creating, verifying or presenting the contents, against specified transfer or exploitation of all or a substantial part.
The protected investment, substantiality of extraction and applicable protection period must be proved. This claim is distinct from ownership of each individual article. [2]
Unfair competition under Articles 54–56 of the Turkish Commercial Code (TTK) is another possible route, although not a separate count in USA TODAY’s complaint. Article 55(1)(c)(3) concerns taking and exploiting another’s market-ready work products through technical reproduction without an appropriate contribution.
A publisher could argue that systematic commercial appropriation satisfies that rule; the provider could contest that characterisation and point to its own technological contribution.
A court would have to establish the statutory unfairness, rather than treating competition for readers or loss of traffic as sufficient in itself. [6]
Evidence would be decisive. FSEK Article 76 permits the court, once the claimant submits evidence sufficient to create a strong conviction concerning its allegation, to require users of protected material to produce authorisation documents or lists of the works used.
Failure to produce the documents or lists creates the statutory presumption of unlawful use. This is a conditional evidentiary mechanism, not an unconditional entitlement to inspect every training record. [2]
Articles 219–220 of the Code of Civil Procedure (HMK) provide for production of relevant documents. Article 400 permits preservation of evidence where the statutory legal interest exists.
An examination should record the model version, query date, full prompt, source article and whether web retrieval was enabled. A RAG answer accessing an article at query time does not, without more, prove that the article was used to train the model.
Under Article 266, experts resolve technical questions; the judge remains responsible for legal characterisation. [7]
Compensation and destruction require separate reasoning
The American complaint invokes exposure up to USD 150,000 for wilful copyright infringement. Under § 504(c), that ceiling concerns the statutory award for an infringed work, subject to the governing conditions; it is not a fresh USD 150,000 award for each copying operation.
Section 1203 provides a separate statutory damages range for § 1202 violations. Turkey has no matching dollar tariff. [1, para 13; 3]
Article 68 FSEK instead permits a claim measured by up to three times the contractual remuneration or the relevant market value for the unauthorised use. This is a licence-based statutory remedy, not an American-style punitive damages award.
Article 70 provides compensation for infringement of economic rights where fault exists and also addresses recovery of profits, requiring deduction of the Article 68 amount when the profit remedy is pursued.
Lost subscriptions, licensing value and attributable profits must therefore be assessed under their respective legal bases, with no duplicate recovery. [2]
Articles 66 and 69 support cessation and prevention; Article 77 allows provisional measures under its statutory conditions. Article 68 expressly provides alternatives including destruction or surrender of unlawfully reproduced copies and relevant reproduction devices.
Model destruction would nevertheless require proof that the particular model falls within the relevant category, as well as analysis of remedial scope and necessity. Article 66 requires consideration of the infringement and the consequences of cessation.
If unfair competition were independently established, TTK Article 56(1)(c) permits destruction of instrumental goods or devices where unavoidable to prevent the unfair competition. [2, 6]
In my assessment, preservation followed by deletion of identified unlawful dataset copies, restrictions on particular outputs or changes to retrieval practices would be more readily supportable than destruction of an entire general-purpose model.
Any such order would still need a sufficient evidentiary and territorial basis. Historical training on a work does not, by itself, establish that destroying the current model is the necessary remedy.
The judgment I would expect on proven equivalent facts
I would expect a Turkish court to examine protection, entitlement, territorial connection and the defendant’s conduct before turning to exceptions and remedies.
If an authorised publisher proved that complete protected articles were copied into a commercial training corpus without permission or an applicable statutory exception, Article 22 would supply a strong basis for relief. The provider could not answer that claim simply by invoking American fair use.
The outcome could nevertheless differ across the claims. The court could uphold the dataset-copying claim while dismissing challenges to summaries conveying only facts or satisfying Article 36.
It could find an Additional Article 4 violation on proven removal of rights information, yet reject claims against group companies whose responsibility was not established.
Substantial database extraction or independently proven unfair competition could support further relief. None of these findings would entitle the publisher to the entire commercial value of the AI service.
Pending legislation does not determine the present dispute
Bill No. 2/3634, submitted on 8 April 2026, proposes amendments to FSEK including a specific licensing mechanism for AI-related uses and certain commercially substitutive outputs.
The official parliamentary record lists it as pending in committee as of this analysis. Proposed Article 42/D therefore cannot be treated as an existing licensing obligation, statutory defence or basis for compensation.
Its relevance is legislative: Parliament is considering a more explicit allocation of rights across the training and output stages. [9]
The present dispute can already be analysed through existing Turkish rights. The strongest claim would identify the protected material, each unauthorised operation and the entity responsible. The strongest defence would address those same elements and the specific exceptions available.
The likely result should follow the proven acts and the corresponding remedies, rather than a general judicial endorsement or rejection of artificial intelligence.
Asutay Duhan Meydan
Attorney at Law
Meydan Law Office