Home About Us Services Awards Team Insights Career Contact Us TÜRKÇEENGLISH Devin Law & IP — Istanbul
← All Insights
CategoryCopyright
Published27 August 2026
Authors
Uğurcan TekinPartner
Beyza ErdemirAttorney at Law

Authorship of AI-Generated Content under Turkish Copyright Law

Generative AI has become an ordinary instrument of business. Advertising visuals, corporate copy, software code and product designs are now produced with these tools, while the legal question has lagged behind the speed of production: is the resulting content a work in the legal sense, and if it is, who holds the economic rights? The answer is commercially decisive. Content that does not qualify as a work enjoys no copyright protection, and unprotected content may be used freely by anyone, competitors included. Being unable to protect the key visual of a campaign or the launch copy of a product is a concrete commercial weakness. This article assesses the current position under the Law on Intellectual and Artistic Works No. 5846, the Turkish Copyright Act.

One point should be stated at the outset. Turkish law contains no statutory regime specific to AI output. The assessment must proceed from the general provisions of the Copyright Act, and those provisions were drafted with human creativity in mind. That is both the source of the problem and the key to its solution.

The line between tool and producer determines the legal outcome. The nature and intensity of the human contribution must be assessed case by case.

The Concept of a Work and the Originality Threshold

The Copyright Act attaches the concept of a work to two conditions (Article 1/B): the content must bear the individual character of its author, and it must fall within one of the statutory categories. The real threshold is individual character, and its definition and scope have been debated intensely in Turkish doctrine. On the broad reading, any intellectual product resulting from a creative activity suffices. On the narrow reading, only products born of serious effort, original in character and capable of establishing a link between the product and its creator cross the threshold. In our view, making protection that easy to obtain damages the balance of interests. Protecting every product that emerges in an ordinary, perhaps accidental way does not encourage true creativity; it obstructs it. For individual character to be accepted, a reasonable perception of a bond between the intellectual product and its creator must arise, and the product must be the result of distinctive work rather than of chance.

The practice of the Court of Cassation gives this framework concrete shape. The settled formula is this: had the work been created by someone else, it would not have been made in the same way and with the same characteristics; individual character is a feature that would emerge only if that particular author made the work (Court of Cassation, 11th Civil Chamber, E. 2020/8509 K. 2022/3996, 24.05.2022; E. 2021/8676 K. 2022/3718, 11.05.2022). The core test, emphasised in case law drawing on Hirsch's classic formulation, is that not everyone could have produced the same work. Quality, by contrast, is not the measure: the product need not be better than its predecessors, for what is sought is the degree of originality, and the assessment is made separately in each individual case. As to authorship, the Act is plain: the author of a work is the person who creates it (Article 8), and Turkish law ties authorship to natural persons. Comparative practice points the same way. The United States Copyright Office has adopted the practice of refusing registration to productions lacking human creativity, and European Union law requires a work to be its author's own intellectual creation.

Individual Character in Practice: Selected Examples

The case law has tested this standard across product types. Maps are an instructive example: every kind of map is listed among the statutory categories, but being listed is not enough to be protected, and whether the specific map bears its maker's individual character must be separately established. In architectural projects the test is even sharper. If five different architects commissioned for the same plot would all draw the same project, there is no work, and only unfair competition protection, where its conditions are met, comes into play. If each architect would draw a different project, individual character and copyright protection exist. Adjusting the door, window and room dimensions of an earlier project does not, by itself, create an independent work.

For slogans, the case law imposes strict conditions: using a common everyday phrase as a slogan creates no individual character, and a close bond of attribution between the slogan and its owner, recognised by the public, is required. Compiling existing information has likewise been held insufficient on its own: in a dispute over lecture slides, technical examination established that the content consisted of pre-existing information, and copyright protection was refused. In musical works the principle of partial individual character applies: a general impression of similarity in the listener's ear is not enough for infringement, and a transfer of the elements carrying individual character is required. The common lesson of these examples is this: protection is earned not by falling within a category but by crossing the individual-character threshold.

AI as a Tool versus AI as the Producer

The working distinction is this. AI may be used as a tool, in the way a camera serves the photographer or drafting software serves the architect. In that case the creative choices belong to the human: the human sets the concept, selects among generated drafts, and edits, combines and finalises the output. Applying the Court of Cassation's test here clarifies the result. Where anyone entering the same prompt can obtain a similar output, the requirement that not everyone could have produced the same work is not met, and no bond forms between the product and the person. Content generated with a single prompt and used as it came will therefore struggle to qualify as a work. Intensive human contribution changes the picture: the activity of selecting, editing and combining personalises the product, a contribution of that intensity can cross the originality threshold, and the person who made it is the author of the resulting work.

Two settled principles offer guidance in this field. The first concerns compilations. Assembling existing information does not, by itself, create individual character; but where the compilation rests on careful effort and carries arrangement choices peculiar to its maker, the threshold can be crossed. The processing of AI output follows the same logic: using the ready-made output as it stands attracts no protection, while the human work of selecting, filtering and recomposing it can create a protectable product. The second principle is partial individual character. Protection exists only for the parts that carry individual character; yet even partial individual character suffices to make the product a work as a whole. In hybrid productions this principle is decisive: the parts carrying human contribution are protected, the purely machine-generated parts fall outside, and the contract and record-keeping architecture must make that distinction provable.

A further layer of the debate is the distinction between fully and semi-autonomous systems, and comparative law recognises it. The United Kingdom's Copyright, Designs and Patents Act 1988 defines a computer-generated work as one generated by computer in circumstances such that there is no human author. Turkish legislation contains no such definition. Since under the present law individual character can arise only in a natural person, fully autonomous output appears incapable of attracting copyright protection. In our view, that position deserves reconsideration in the face of advancing technology, and granting protection to fully autonomous productions through an exceptional statutory provision would better suit present conditions. In semi-autonomous production, by contrast, the human influence is high both in writing and in applying the code, and individual character should there arise only in the natural person concerned. This distinction is the statutory-level counterpart of the tool-versus-producer distinction described above.

An example makes the originality debate concrete. Ask a person twice, with the same instruction, to draw a purple baby elephant, and the two drawings will differ: the grip of the pen, the lines and the colouring change each time, and creativity enters at precisely that point. Give the same prompt twice to an AI system and it stays, as a rule, within the variety its algorithm permits. Considering that AI cannot form a declaration of intent and operates within the boundaries its creator drew, the conclusion is that individual character in the classical sense does not arise in the machine. At the same time, the limits of the human brain's own capacity for variation are not fully known either, and categorically denying AI any capacity for originality is, in our view, equally unsound. The debate is open at the level of legal policy and awaits resolution by statute.

Proving which side of the line a given output falls on is a matter of records. Which prompts were given, how many drafts were generated, which selections were made and what changes were carried out by hand should all be documented. Those records will be the principal evidence in any future originality dispute. An undocumented production process remains unprovable, however intensive the contribution actually was.

“In AI output, the insurance policy of ownership is the documentation of the production process. Where there is no record, the contribution counts for nothing.”

Practical Consequences for Companies

The first consequence concerns work product. The economic rights in works created by employees in the performance of their duties are, as a rule, exercised by the employer (Article 18 of the Copyright Act). But that presumption operates only where a work exists; unprotectable content produced by an employee with a single prompt never enters it. The second consequence concerns client relationships. Agencies and software developers pass their work to clients by assigning economic rights, and where the AI-generated part of the deliverable is unprotectable, the subject matter of the assignment itself becomes doubtful. Declaring the use of AI in service agreements has become important for precisely this reason.

The third consequence is infringement risk. Generative models are trained on existing content, and it is entirely possible for an output to resemble a third party's work. When resemblance turns into an infringement claim, liability as a rule stays with the party that used the content. The risk must be managed contractually: content procured from suppliers should carry warranties as to third-party rights, and for critical campaigns the originality of the output should be separately checked before publication.

Contract and Policy Recommendations

Corporate use should rest on a written policy. The policy should determine which tools may be used for which work, and the entry of confidential information and client data into such tools should be separately regulated. Documentation of production processes should be mandatory. Client agreements should state expressly the scope of the AI contribution and the limits of the rights assignment. For assets whose copyright status remains uncertain, alternative layers of protection should be considered: trademark registration can protect slogans and logos, design registration can protect product appearances, and unfair competition rules offer an additional ground against the copying of work product.

A Checklist

  • Put AI use on a written corporate policy, and restrict the entry of confidential information and personal data into the tools.
  • Document the production process: record prompts, drafts, selections and the changes made by hand.
  • Keep the human contribution at the centre of production, and do not take single-prompt output into commercial use as it stands.
  • Add to client and supplier agreements an AI declaration, the scope of the rights assignment and warranties covering third-party rights.
  • For assets whose copyright status is uncertain, consider trademark and design registration and the protection of unfair competition law.

The legal regime of AI output is not yet settled: there is no statute, and judicial practice is still forming. In this environment of uncertainty, what protects a company is cautious structuring. Documenting the human contribution, drafting agreements in plain terms and activating alternative layers of protection are the measures available today. The uncertainty will one day be resolved by case law or by statute. The company that meets that day prepared is the one whose records and contracts are already in order.

Sources

  • Law on Intellectual and Artistic Works No. 5846, Articles 1/B, 8 and 18; Copyright, Designs and Patents Act 1988 (UK) on computer-generated works
  • Court of Cassation practice cited: 11th Civil Chamber, E. 2020/8509 K. 2022/3996, 24.05.2022; E. 2021/8676 K. 2022/3718, 11.05.2022; E. 2020/5691 K. 2021/6489, 23.11.2021; E. 2021/2610 K. 2022/6700, 05.10.2022; E. 2012/11315 K. 2014/4768, 12.03.2014; General Assembly of Civil Chambers, E. 2017/7 K. 2020/185, 20.02.2020
  • Doctrine: Hirsch, Hukuki Bakımdan Fikri Say, Vol. II, Istanbul 1943; Arslanlı, Fikri Hukuk Dersleri II, Istanbul 1954; Ateş, Fikri Hukukta Eser, Ankara 2007; Yavuz/Alıca/Merdivan, Fikir ve Sanat Eserleri Kanunu Yorumu, Seçkin; Çekin, Yapay Zekâ Teknolojilerinin Hukuki İşlem Teorisine Etkileri; Davies, An Evolutionary Step in Intellectual Property Rights, Computer Law & Security Review
  • mevzuat.gov.tr: current texts of the cited statutes
  • United States Copyright Office: announcements and registration guidance on AI-generated material
Related practice areaIntellectual Property & Copyright Law→

Related reading

CopyrightProtection of Unregistered Copyright Against Trademark Applications14 February 2026IP LitigationPreliminary Injunctions and Evidence Preservation in Turkish Trademark Infringement Litigation27 August 2026TrademarkThe Quietest Way to Lose a Right: Renewal Calendars and Bulletin Watch26 August 2026