Suno copyright law in Turkey raises a number of unresolved questions concerning the use of protected music in generative AI systems. As litigation against Suno develops across the United States, Germany and Canada, the legal debate has expanded beyond model training to encompass the acquisition of training data, phonogram producers’ rights, model memorisation, generated outputs and the protection of artists’ identities.

A Legal Analysis of Generative AI Music Platforms under Articles 80, 76 and 68 of the Turkish Law on Intellectual and Artistic Works

Av. Asutay Duhan Meydan
Meydan AI Law
5 September 2026

What is most striking about the litigation now unfolding against Suno is no longer simply the number of cases being filed. It is the way in which the legal character of the dispute itself is changing.

When major record companies commenced proceedings against Suno in the United States in 2024, the core allegation was relatively familiar from a copyright perspective: had copyright-protected sound recordings been reproduced without authorisation for the purpose of training an artificial intelligence model?

Within two years, the dispute has moved well beyond that question.

In the Massachusetts proceedings brought by UMG, Capitol Records and Sony Music Entertainment, the manner in which Suno obtained its training material has itself become a central issue during discovery. According to court filings, Suno disclosed that audio files had been downloaded from YouTube using open-source tools such as YT-DL and YT-DLP. The plaintiffs subsequently sought to introduce additional claims under the DMCA, alleging that YouTube’s technical mechanisms restricting direct access to those files had been circumvented.

In its decision of 18 August 2026, the court held that the circumvention theory was, at the very least, sufficiently pleaded to proceed, while leaving the ultimate question of whether the relevant measure constituted an effective technological access control for determination at a later evidentiary stage.

The action filed by Round Hill Music against Suno and Bright Data on 17 August 2026 reflects the same development. The case is not limited to allegations of unauthorised reproduction. It also includes theories concerning contributory infringement, circumvention of technological protection measures, and the removal or alteration of copyright management information.

Gerencia 360 Music likewise commenced separate proceedings against Suno and Bright Data on 31 August.

The Canadian action brought by SOCAN on 2 September 2026 focuses on yet another aspect of the problem: whether outputs generated by Suno reproduce musical works within SOCAN’s repertoire and whether making those outputs available through the platform itself gives rise to further infringement.

Proceedings brought in the same period by artists including David Lowery and Jason Isbell open a different legal front altogether. There, the central issue is no longer ownership of the sound recording itself, but the commercial exploitation of an artist’s name, voice, performance characteristics and personal identity.

Alongside these cases, the judgment delivered in GEMA v. Suno by the Munich I Regional Court on 31 July 2026 is, for the time being, one of the most significant European developments. The court substantially upheld GEMA’s claims for injunctive relief, information and damages. It found that six musical works at issue were reproducible from certain versions of Suno’s models in a manner capable of amounting to legally relevant reproduction, and further concluded that certain generated outputs reproduced protected elements of those works.

The judgment is not yet final. Nevertheless, it is significant as one of the first substantive European decisions to examine generative AI music systems at this level of detail.

Accordingly, the Suno litigation can no longer be reduced to the single question of whether copyrighted music may lawfully be used for AI training.

At least four distinct stages must now be separated:

the acquisition of the data, the training of the model, the retention or memorisation of protected material within the model, and the outputs made available to users.

Any action brought in Türkiye should make the same distinction.

Who should bring the action in Türkiye?

The first mistake in a Turkish action would be to assume that a record company is automatically the “copyright owner” of every song in its catalogue.

A single music recording may contain several legally distinct layers of rights.

The composer and lyricist hold rights in the musical work. The performing artist enjoys separate related rights. The producer responsible for the first fixation of the recording is protected independently as the phonogram producer.

For a record company, the most secure starting point will therefore often be its status as a phonogram producer under Article 80 of Law No. 5846 on Intellectual and Artistic Works (“FSEK”).

Article 80 grants the phonogram producer exclusive rights in relation to, among other things, the direct or indirect reproduction and distribution of the phonogram and its communication to the public by means capable of transmitting signs, sounds or images.

This distinction is fundamental to establishing standing correctly.

The fact that a record company produced an artist’s master recording does not automatically mean that it owns all economic rights in the underlying composition and lyrics. Likewise, an artist’s personality rights in his or her voice do not form part of the record company’s assets merely because the company owns rights in the master.

Before proceedings are commenced, therefore, the catalogue must be subjected to a proper chain-of-title review.

For each recording, it should be established separately:

who qualifies as the phonogram producer;
who owns the rights in the composition and lyrics;
what rights have been granted under the artist agreements; and
whether digital exploitation and communication-to-the-public rights have been transferred.

A strong first Suno case in Türkiye should not begin with a catalogue of one hundred thousand songs.

It should begin with a limited number of phonograms in respect of which the claimant’s title is beyond serious dispute.

Which exclusive right could AI model training infringe under Turkish law?

FSEK does not currently contain a provision expressly referring to the training of artificial intelligence models.

That does not mean that model training falls outside the scope of the existing copyright framework.

Article 22 of FSEK defines the reproduction right in broad and technologically neutral terms. The reproduction of a work, in whole or in part, directly or indirectly, temporarily or permanently, by any means or method, falls within the exclusive right of the copyright owner. Recording a work on devices used for the transmission of sound or images likewise constitutes reproduction.

For a phonogram producer, the same issue arises through the direct and indirect reproduction right protected under Article 80.

During the preparation of a recording for model training, several technically distinct acts may occur: downloading the file, storing it on a server, incorporating it into a training dataset, preprocessing it, or creating temporary or intermediate copies during training.

Depending on the technical facts of the particular case, any of these acts may engage the reproduction right.

However, one point is essential:

It is not enough to say: “The model was trained, therefore FSEK must have been infringed.”

The technical act must be identified and put before the court.

Which phonogram was obtained? From where? Was it copied onto a server? Was it included in the training corpus? Which model version used it? Following training, does the model retain protected elements of the recording in a reproducible form?

This is one reason why the GEMA judgment is particularly significant.

The Munich court did not merely reason that “AI learns, therefore AI copies.” It examined whether the relevant works could be reproduced from particular model versions and treated the question of model memorisation as legally relevant.

There is, of course, no basis for assuming that a Turkish court would necessarily reach the same conclusion.

The significance of the GEMA judgment lies elsewhere: it demonstrates that the proposition is no longer merely academic.

Suno cannot simply import the US “fair use” defence into Turkish law

One of the principal arguments available to defendants in US litigation is that the use of protected material for model training is transformative and should therefore qualify as fair use.

Turkish law contains no general equivalent to §107 of the US Copyright Act that would permit courts to apply an open-ended, multi-factor fair-use doctrine.

The limitations and exceptions to exclusive rights under FSEK are instead regulated through specific statutory provisions.

Nor does Turkish law currently contain a general exception authorising commercial AI training under a broad text and data mining rule.

This distinction may place a Turkish record company in a materially different substantive position from a claimant litigating in the United States.

It does not, however, resolve the case.

In Türkiye, the most difficult issue may arise before the merits of any “fair use” type argument are even reached:

territoriality and applicable law.

The central conflict-of-laws question: what if the training took place in the United States?

There is an important shortcut that should be avoided.

A claimant cannot simply argue that every act of reproduction carried out on Suno’s servers in the United States infringes FSEK merely because proceedings have been brought before a Turkish court.

Article 23 of Law No. 5718 on Private International Law and International Civil Procedure provides that intellectual property rights are governed by the law of the country for which protection is sought.

This reflects the principle commonly described as lex loci protectionis.

The individual acts must therefore be separated.

If training material was acquired in the United States and the relevant training copies were also created there, the law governing those acts must be determined separately.

By contrast, the availability of Suno’s system in Türkiye, the reproduction of protected material at model or output level within Türkiye, its communication to users in Türkiye, or other effects upon rights protected in Türkiye may require a different legal analysis.

For that reason, the correct theory of the case cannot simply be:

“Suno trained its model in the United States and therefore infringed Article 22 of FSEK.”

The correct question is:

Within the chain consisting of data acquisition, model training, retention within the model, output generation and the provision of services in Türkiye, which acts are alleged to infringe rights protected in which jurisdiction?

The GEMA judgment is instructive in this respect as well.

The German court distinguished between training copies made in the United States and conduct relating to the model and outputs in Germany, considering the US training activity separately under US law.

That approach cannot simply be transplanted into Turkish law, but it illustrates the manner in which a cross-border Suno dispute must be disaggregated.

As to international jurisdiction, Article 40 of Law No. 5718 refers to the domestic rules on territorial jurisdiction.

For tortious conduct, Article 16 of the Turkish Code of Civil Procedure (“HMK”) recognises jurisdiction, inter alia, at the place where the wrongful act occurred, where the damage occurred or is likely to occur, and at the domicile of the injured party.

There is therefore a credible basis for asserting the international jurisdiction of Turkish courts in respect of infringement and damage alleged to arise in Türkiye.

A foreign defendant such as Suno should nevertheless be expected to raise serious objections concerning both jurisdiction and applicable law.

What type of action should be brought in Türkiye?

For a record company, I would not frame the case as a simple damages action.

Its core should instead consist of:

a declaration of infringement of the phonogram producer’s related rights, cessation and prevention of infringement, disclosure and production of documents under Article 76 FSEK, and monetary relief under Article 68 FSEK.

Depending on the facts and evidence, the relief sought could be structured broadly as follows:

Declaration of rights and infringement
A declaration that the claimant holds the related rights of a phonogram producer under Article 80 in respect of the identified recordings, and that Suno’s unauthorised use of those phonograms constitutes infringement.
Cessation and prevention of infringement
Relief under Articles 66 and 69 of FSEK requiring the cessation of existing infringement and preventing further use of the identified phonograms for new model training, fine-tuning or infringing output generation directed at Türkiye.
Production of records under Article 76
An order requiring Suno to produce information and documentation identifying whether the specified phonograms were used in any model version, from which sources they were acquired, and whether any licence existed for such use.
Interim injunctive relief and preservation of evidence
Measures under Article 77 FSEK and Article 389 HMK preventing the deletion, alteration or destruction of training-data records, source information, model-version records and other relevant technical material while proceedings are pending.
Monetary relief under Article 68 FSEK
Recovery, in respect of past unauthorised use, of the amount that would have been payable had a licence been concluded, or the applicable market royalty, subject to the statutory multiplier permitted by Article 68.
Further damages or disgorgement under Article 70, where appropriate
Claims for loss or surrender of profits may also be considered, provided they are structured so as to avoid double recovery in respect of the same economic harm.

Two matters deserve particular emphasis.

First, the initial interim application should not be framed as:

“Block Suno entirely in Türkiye.”

Such relief would unnecessarily expose the application to serious objections based on proportionality and the evidentiary threshold applicable to interim measures.

If the case concerns twenty phonograms, the interim relief should, as far as practicable, be tailored to those twenty phonograms.

Secondly, relief under Article 68 and claims for cessation or prevention should be structured carefully in temporal terms.

Turkish case law has treated payment under Article 68 as giving rise, in respect of past use, to a form of hypothetical contractual relationship. That may affect the availability or scope of cessation and prevention orders relating to the same past use.

A cleaner structure is therefore to distinguish between:

monetary relief under Article 68 for completed past infringements; and
preventive relief under Article 69 in respect of continuing or future uses.

Article 76 FSEK may become the key provision in a Turkish Suno case

The first evidentiary problem facing a Turkish record company is obvious:

How can the claimant prove that its recordings were included in Suno’s non-public training corpus?

This is not a theoretical difficulty.

Training datasets used by major AI companies are frequently inaccessible to outsiders. Requiring a claimant to prove the defendant’s entire internal technical architecture before commencing proceedings could, in practical terms, deprive the right holder of any effective remedy.

Turkish law contains a particularly important mechanism in this respect.

Where the claimant produces evidence sufficient to create a strong judicial conviction as to the plausibility of the infringement allegation, Article 76 FSEK permits the court to require the user of protected works or phonograms to produce the necessary licences and authorisations, or lists identifying the works and phonograms used.

Failure to produce the relevant documentation may give rise to a statutory presumption of unlawful use.

This provision should not be confused with US-style discovery.

Article 76 does not permit a claimant with no evidentiary foundation to demand disclosure of Suno’s entire global training corpus.

The claimant must first satisfy its own initial evidentiary burden.

Turkish courts have likewise approached attempts to infer infringement of an entire repertoire from a limited number of proven uses with caution.

The proper sequence in a strong Suno case should therefore be:

first, prima facie evidence; then Article 76.

Suppose a record company conducts controlled testing on twenty or thirty phonograms and obtains several strong examples of reproduction.

Suppose further that a technical expert concludes that the relevant similarities cannot readily be explained by coincidence, common musical vocabulary or mere stylistic resemblance.

If public statements and disclosures from international Suno proceedings concerning the acquisition of training material are then also placed before the Turkish court, an Article 76 request begins to look materially different.

It is no longer a speculative fishing expedition.

It becomes a targeted request for information that is both relevant to specifically identified recordings and largely within the exclusive control of the defendant.

Before filing the merits action, the first priority may be to preserve the evidence

An output generated by a continuously changing AI system today may not be reproducible using the same prompt three months later.

The model version may change.

Safety filters may change.

The output may be removed.

An account may be suspended.

A work that was memorised by one version of the model may no longer be reproducible from a later version.

For this reason, the first legal step may not necessarily be the substantive action itself.

In the right case, preservation of evidence under Article 400 HMK may prove more valuable.

That application, however, should not be framed abstractly as:

“Determine whether Suno is infringing copyright.”

The purpose of evidence preservation is not to obtain an advance determination of the merits.

The applicant must identify concrete evidence that is at risk of disappearing or becoming materially more difficult to produce at a later stage.

A technical testing protocol should therefore be established first.

A representative set of recordings in respect of which the claimant’s title is undisputed should be selected.

For each test, the following should be retained:

the exact prompt used;
the date and time;
the relevant Suno model version, where identifiable;
the account information;
the original output file; and
a cryptographic hash of that file.

The original phonogram and the output should then be compared by experts in musicology, audio engineering and, where necessary, machine learning.

The relevant legal inquiry is not simply whether “the two songs sound similar.”

Musical genre, harmonic conventions or general mood are not, by themselves, protected subject matter.

The real question is whether specific and legally protectable elements of the claimant’s phonogram or musical work have been reproduced.

Without that distinction, the case can quickly collapse into an argument over whether an artist may monopolise a musical “style.”

There is also a practical contractual issue.

Suno’s current Terms of Service exclude intellectual property infringement disputes from the definition of arbitrable disputes, but include a Massachusetts forum-selection clause for disputes not subject to arbitration.

Whether that clause can bind a phonogram producer that has never entered into any contractual relationship with Suno is a separate question.

Nevertheless, if an account is casually opened in the claimant company’s name merely for evidence-gathering purposes, Suno may at least attempt to rely upon those contractual provisions.

Even the identity of the person or entity operating the testing account, and the legal capacity in which it is used, should therefore form part of the pre-litigation strategy.

How should damages be calculated? This may become the real economic battleground

Article 68 FSEK enables a right holder, where a work, performance or phonogram has been reproduced or otherwise exploited without the required written authorisation, to claim the amount that would have been payable had a licence been concluded, or the applicable market value, subject to the statutory enhancement mechanism.

In a Suno case, the difficult question will not merely be whether Article 68 applies.

The more important question will be:

What is the relevant market royalty?

The answer should not simply be derived from Spotify streaming revenue or an ordinary digital distribution licence.

The use is fundamentally different.

The hypothetical licence should instead be framed as follows:

What would a technology company have paid to license a defined catalogue of Turkish music for the purpose of training, developing and improving a commercial generative AI music model?

The expert valuation may therefore need to consider:

the size of the catalogue;
the commercial value of the relevant phonograms;
the number of recordings actually used;
the duration of the use;
the economic scale of the model;
the number of users and subscribers;
whether the recordings were repeatedly used across training cycles; and
whether the generated outputs have a substitution effect upon existing licensing markets.

Interestingly, Bill No. 2/3634, currently pending before the Turkish Grand National Assembly, contemplates precisely this type of consideration for future AI licensing tariffs, including the intensity of use, volume of data, commercial reach of the model, economic scale of the service and value of the outputs.

The bill is not currently in force and cannot constitute the legal basis of a claim brought today.

It is nevertheless highly relevant as an indication of the direction in which the Turkish legislative debate is moving.

Claims under Articles 68 and 70 must also be coordinated so that the claimant does not recover twice for the same economic loss.

The better approach is therefore to plead alternative forms of monetary relief, or expressly provide for set-off between overlapping heads of recovery.

If the artist’s voice is being imitated, the record company’s copyright action may not be enough

The second branch of the Suno problem exists independently of copyright.

A system may not reproduce a record company’s master recording, yet may imitate a particular artist’s voice, vocal characteristics or other distinctive elements of personal identity in a commercially exploitable manner.

Here, the record company’s rights under Article 80 FSEK must not be confused with the artist’s personality rights.

Türkiye already has a significant appellate decision touching upon this issue.

In its decision dated 22 May 2026, Ankara Regional Court of Appeal, 20th Civil Chamber, File No. 2026/1078, Decision No. 2026/1119, considered a dispute involving the processing of a well-known individual’s voice through artificial intelligence and an application that allowed users, for payment, to make the voice utter requested sentences.

The Chamber held that, as pleaded, the dispute fell primarily within the framework of personality rights and data protection rather than FSEK, and accordingly upheld the conclusion that the Civil Court of First Instance had jurisdiction.

The judgment is not a substantive ruling that “AI voice cloning is unlawful.”

It is a jurisdictional decision.

Nevertheless, it is highly significant because it indicates how Turkish courts may begin to classify such disputes.

Accordingly, where a Turkish artist’s voice is recognisably imitated by Suno, a separate claim may potentially be developed under:

Articles 24 and 25 of the Turkish Civil Code, Article 58 of the Turkish Code of Obligations, and, depending on the nature of the processing, the Turkish Personal Data Protection Law.

It would be more coherent to maintain a deliberate distinction between the record company’s FSEK claim concerning the master recording and the artist’s personality-rights claim, rather than treating them as the same cause of action.

The bill before Parliament attempts to regulate this very problem

The Turkish debate has become even more significant following the submission of Bill No. 2/3634, proposing amendments to FSEK.

As of 5 September 2026, the bill remains before committee and is not in force.

The proposed Article 42/D would establish a licensing framework for the reproduction, processing, extraction, data-mining and storage of works, performances and phonograms for the purposes of:

training AI systems;
developing them;
fine-tuning them;
testing and evaluating them;
creating datasets; and
improving model performance.

The proposal would also introduce a separate licensing framework for certain commercial uses of AI-generated outputs.

Its significance for present litigation is straightforward:

A claimant does not need to wait for the bill to become law before bringing proceedings against Suno in Türkiye.

A claim today may already be developed under Articles 22, 66, 68, 69, 70, 76 and 77 FSEK and, in the case of a phonogram producer, particularly Article 80.

The proposed legislation nevertheless confirms something important at policy level: the Turkish legislature itself has identified the use of works and phonograms for AI training as a distinct licensing and rights-management problem.

What should Türkiye’s first Suno case look like?

In our view, the first Turkish case capable of generating meaningful precedent should not be a sweeping action alleging that Suno has “stolen Turkish music” as a whole.

That may appear rhetorically powerful, but it imposes an unnecessary evidentiary burden.

A stronger test case would be built around ten to twenty commercially valuable and technically distinctive master recordings in respect of which the claimant’s chain of title and status as phonogram producer are clear.

Those recordings should be subjected to controlled testing.

A small number of strong reproduction examples should be identified.

The resulting evidence should be preserved before the system changes.

Only then should the court be asked—not to order disclosure of Suno’s entire global training corpus—but to require production, under Article 76, of the records showing:

from which sources the identified phonograms were obtained;
in which model versions they were used; and
what licences, if any, authorised such use.

At that point, the claimant’s case would no longer rest on speculation.

It would be able to say to the court:

“We own the rights in these phonograms. We did not license them to Suno. Controlled testing of the system has produced concrete results requiring explanation by reference to protected material. Suno is the only party with access to the relevant training-data and licensing records. We therefore seek production of those records under Article 76 FSEK, cessation of the unauthorised use, and payment of the licensing value of the past exploitation.”

In our view, this is the legal core of the most credible Suno action presently capable of being brought in Türkiye.

Conclusion: the problem is not that Turkish copyright law cannot reach Suno; it is that the boundaries have not yet been drawn by the courts

Turkish law does not currently recognise a special cause of action called an “AI training claim.”

Nor is one necessarily required.

The existing provisions of FSEK are capable, in an appropriate case, of bringing before the courts questions concerning:

reproduction occurring during AI training;
the related rights of phonogram producers;
cessation and prevention of unauthorised use;
hypothetical licence fees;
disclosure of relevant records; and
preservation of evidence.

The real uncertainty lies elsewhere.

Turkish courts have not yet answered when a model should be regarded as “using” a phonogram, when model memorisation amounts to reproduction, where the territorial consequences of foreign training activity begin for purposes of Turkish law, what evidence is sufficient to connect an output to training material, or how the market value of an AI-training licence should be calculated.

For that reason, the first serious Suno case in Türkiye would not merely be a damages action brought by a record company.

It would place a broader question before the Turkish courts for the first time:

Does the economic value of a music catalogue arise only from human consumption of the music, or does the use of that catalogue to create a commercial artificial intelligence model capable of learning from it and generating new music carry a separate licensing value of its own?

In our view, the answer to that question is likely to become one of the defining issues of Turkish copyright law in the age of artificial intelligence.

This article has been prepared on the basis of Turkish law in force as of 5 September 2026 and publicly available foreign court records and judgments. Allegations made in pending foreign proceedings should not be treated as established facts or judicial findings unless expressly stated otherwise.