Music Generated with AI and Copyrights

Music Generated with AI and Copyrights. Artificial intelligence-based voice cloning technologies (such as SUNO) have rapidly transformed the music industry. High-quality synthetic voices are being produced while preserving accents and nuances. This technological advancement has brought with it legal challenges. Traditional copyright regimes are inadequate to address this issue. Therefore, the legal battle has shifted to the area of personal rights, protecting artists’ voice and identity rights. Internationally, the US’s “Fair Use” approach conflicts with the EU’s “Opt-out” regulation. In Türkiye, the Law on Intellectual and Artistic Works (FSEK) provides fundamental protection. In this climate of uncertainty, the industry has turned to controlled licensing models, thus seeking a balance between AI and creativity.

I. Executive Summary and Strategic Outlook

The rapid advances made by Generative Artificial Intelligence (AI) technologies in the fields of music and audio production are fundamentally shaking global intellectual property (IP) regimes. AI voice cloning possesses the capability to create high-quality customized audio content, even preserving the accent and nuances of an individual’s voice. However, this technological capacity poses serious threats, particularly to artists’ copyrights, performers’ rights, and personality rights.

Analyses indicate that legal protection against AI voice cloning is finding a more effective footing through personality rights (identity protection) rather than through traditional copyright law (protection of the work). Specifically, in Turkey, the provisions on Moral Rights within the Law on Intellectual and Artistic Works (FSEK) stand out as the strongest defense mechanism for protecting the reputation and integrity of the artist’s work. Meanwhile, globally, a deep regulatory divide (global crack) exists between the US “Fair Use” approach regarding the training of AI models and the European Union’s (EU) “Opt-out” model, which grants rights holders the right to object.

Industry leaders are undergoing a strategic transformation in response to legal uncertainty. Major record labels, such as Universal Music Group (UMG), have opted to maintain market dominance by converting their legal battles with AI developers (like Udio) into controlled licensing partnerships that ensure artist control and royalty payments. This approach signals that the future of the music industry will be shaped by sectoral business models and transparency protocols in an environment where laws lag behind technology.

II. Fundamentals and Evolution of Generative Audio Technology

A. AI Voice Cloning Mechanisms and Quality Standards

Deep Learning models form the basis of AI voice cloning technology. These models are capable of producing high-quality synthetic audio content that can mimic speech (Text-to-Speech – TTS) or singing voices. The cloning process begins with the uploading or recording of voice data (sometimes only one minute of recording may suffice) into AI systems like Genny. The model analyzes this data and learns the fundamental characteristics of the original voice.

At the current stage of technology, the produced voice clones not only provide accurate pronunciation but also preserve the accent and nuances of the original speaker. The preservation of these subtle details is of great importance in the context of legal disputes. This is because the targeted imitation of distinctive features—which are part of the artist’s commercial and artistic identity—by AI systems, rather than generating a general voice profile, can be accepted as strong evidence supporting claims of personality rights (Right of Publicity) infringement.

Singing Voice Conversion (SVC)

Singing Voice Conversion (SVC), a subfield more complex than speech cloning, offers great possibilities in the music domain. SVC technology requires not only the conversion of speech characteristics but also the mapping of musical elements like melody and rhythm to the target voice. AI singing voice generators like So Vits SVC have the capacity to take a source singing voice and convert it into a targeted AI singing voice, accompanied by the lyrics, harmony, and rhythm.

B. Text-to-Music Generation Models

The history of AI music production began with algorithmic composition methods and computer-generated music in the mid-20th century. With the introduction of neural networks in the 1990s, style imitations were produced by analyzing and copying the styles of famous composers, as seen in works like David Cope’s EMI study. Today, deep learning technology (Google’s Magenta project and OpenAI’s MuseNet) has made more sophisticated music generation possible.

The competition in this field is intensifying. OpenAI is collaborating with Juilliard School students to develop an AI tool capable of generating music from text and voice commands, creating data sets to train the AI. The company’s goal is a system that can offer new creative avenues, such as generating guitar accompaniment for a vocal track or adding original music to videos. However, these developments, along with the launch of tools by startups like Suno and ElevenLabs, have also raised concerns about the proliferation of AI-generated “spam” music on digital platforms.

The tendency of AI algorithms to prioritize popular trends and formulas in their training data may reduce the diversity of artistic expression. This situation raises ethical concerns that algorithmic systems could contribute to the homogenization of musical styles and lead to the marginalization of niche genres. This demonstrates that AI music production simultaneously carries the potential to increase creativity and the imperative to protect musical diversity.

III. Copyright in AI-Generated Works

A. Concept of Work and the Human Labor Requirement under FSEK

Both the Turkish Law on Intellectual and Artistic Works (FSEK) and international copyright law fundamentally protect only “original, human-authored” works. FSEK does not recognize abstract ideas that cannot be perceived by third parties as works; a creation must be embodied in a specific form to benefit from protection.

In legal assessments, entirely AI-generated outputs are denied copyright protection because human contribution is not found at the core of the creation. The US “Zarya of the Dawn” decision reinforced this principle, viewing AI as merely a creative tool and granting copyright to the human creators who utilize that tool. This entails the risk of autonomously generated works remaining “unclaimed” and passing into the public domain. Therefore, for AI-generated music to be protected, creative contributions made by humans in the creation process—such as selections, arrangements, or commands—must be clearly evidenced.

B. Limitations of Copyright Protection over Sound Recordings

In the legal struggle of voice cloning technology against copyright, federal copyright law has been observed to be insufficient. Court rulings in the US have clearly stated that copyright law protects only the original sound recordings (the fixed expression), but not the abstract qualities of a voice or new imitation recordings.

An AI-generated voice clone, no matter how perfectly it imitates the original recording, is technically considered a new and independent fixation (fixation). Therefore, the output produced through cloning may not constitute copyright infringement (even derivative work infringement) unless it is a direct reproduction of the original recording. When courts conclude that AI-generated outputs do not constitute direct infringement, claims of contributory infringement are also dismissed.

This situation demonstrates that the legal struggle is shifting from the field of intellectual property (IP) to the law of personality and unfair competition, which protects the identity and reputation of the individual. The inability of federal copyright protection to protect the abstract quality of the voice forces artists and rights holders to turn to alternative legal remedies to combat cloning.

IV. Protection of Voice as a Personality Right and Infringement

A. Legal Nature of Voice within the Scope of Personality Rights

Voice is a unique intangible asset used for communication, identity verification, and artistic expression in human society. Especially for public figures and celebrities, the voice is protected as a valuable asset under the Right of Publicity, which is the authority to control the commercial use of one’s name, image, and likeness.

The unauthorized use of a cloned voice to promote or endorse products or services leads to serious legal consequences. In international litigation, the main distinguishing point of personality rights infringement is whether the cloned voice was used for a commercial endorsement purpose. If the cloning involves commercial identity impersonation, the claims are strengthened. Furthermore, cloned voices can be misused for identity fraud and scams to gain access to personal or financial information, leading to criminal activities.

B. International Case Examples

In US lawsuits, plaintiffs (e.g., Lehrman and Sage/Lovo case) have concluded that they could be more successful through state-law-based personality rights (right-of-publicity) remedies against AI cloning, rather than federal copyright protections. The court ruled that the use of a cloned voice does not satisfy the false association elements of the federal Lanham Act unless the voice is used to directly endorse or promote a product or service.

In Asia, the lawsuit filed by voice actress Yin against Beijing Intelligent Technology Company (BCT) in China serves as a precedent. Despite BCT owning the copyright to the original recordings, it used Yin’s voice in AI text-to-speech (TTS) products without authorization. The court ruled that owning the original recordings did not eliminate the requirement for additional permission to use the voice itself in an AI product, finding that Yin’s IP/Right of Publicity was infringed.

C. Protection of Voice under KVKK in Turkey

In Turkish Law, an individual’s voice is treated as personal data under the Law on the Protection of Personal Data (KVKK). The recording and sharing of voice are subject to KVKK provisions. The law restricts the processing and sharing of voice recordings without the explicit consent of the data subject. This provision makes obtaining the consent of voice owners an absolute necessity in the training of commercial AI voice cloning models or the distribution of their outputs to users.

In Turkey, the unauthorized cloning of an artist’s voice carries the potential to constitute an infringement in two separate legal areas: both KVKK (infringement of consent for voice data) and FSEK (infringement of Moral Rights). The risk of administrative fines arising from KVKK constitutes a significant deterrent for companies and forces AI models toward consent-based training.

V. Critical Area: AI Training Data and Copyright Infringement

A. Training on Copyrighted Datasets: Global Challenges

Generative AI models must be trained on vast datasets protected by copyright (Text and Data Mining – TDM) to maximize their performance. The unauthorized and uncompensated use of copyrighted music to train these models has led to a legal impasse between rights holders and AI developers. This situation has caused major record labels to file lawsuits against AI producers for infringement.

B. International Comparison: US Fair Use vs. EU Opt-Out Model

Two main global approaches exist regarding the use of AI training data:

US Approach (Fair Use): In the United States, the use of copyrighted material for AI training is generally assessed under the “fair use” doctrine. This defense may not require obtaining permission from rights holders under certain conditions.
European Union (EU) Regulations (Opt-out): The EU has established a specific framework for AI model training through the Digital Single Market Directive and the EU AI Act.
- Non-Commercial Use: Research organizations and cultural heritage institutions can use works to which they have lawful access for TDM for scientific research purposes, and rights holders cannot object.
- Commercial Use: In commercial or non-research TDM applications, copyright holders can prevent the mining of their works by making an express reservation of right (Opt-out) in an appropriate manner. This Opt-out model is considered a compromise favoring rights holders.

The EU AI Act also imposes significant transparency obligations on providers of general-purpose AI models. This requires the publication of a detailed, comprehensive summary of the content used to train the model. Crucially, these obligations apply to all models placed on the EU market, regardless of where the training took place, carrying the risk of extraterritorial application.

The withdrawal of Getty Images’ fundamental copyright infringement claims against Stability AI in the UK due to jurisdictional hurdles demonstrated how central the issue of jurisdiction is in the legal struggle over AI training data. This indicates that AI companies will tend to choose less restrictive legal jurisdictions, such as “fair use” in the US, while the EU’s Opt-out model is a proactive regulatory step aimed at closing this jurisdictional gap.

Table 1: Copyright Regulations in AI Training (US vs. EU Comparison)

Regulatory Area	US (Fair Use Doctrine)	EU (Digital Single Market Directive & AI Act)
Default Approach	Permission may not be required with the Fair Use defense.	Permission or Opt-out possibility by the rights holder is essential for commercial use.
Data Mining (TDM) Restriction	Determined by high court rulings (still evolving).	Rights holders have the explicit right to object (Opt-out).
Transparency Obligation on AI Provider	Limited and Developing.	Obligation to publish a comprehensive summary of training content (High).
Risk of Extraterritorial Application	Low.	High (Applies to all models entering the EU market).

VI. Voice Cloning and FSEK Protection in Turkish Law

A. Related Rights Ownership for Sound Recordings and Performances under FSEK

AI voice cloning has the potential to infringe upon the economic rights (pecuniary rights) of performing artists over their work. These pecuniary rights include the rights of reproduction, distribution, and communication to the public by means of signal, sound, and/or image transmission. FSEK protects the related rights of performing artists over their performances recorded by means of sound and image fixation tools.

However, the current trend in international law (see Section III.B) makes it difficult for this to be seen as a direct infringement of reproduction or pecuniary rights, as the AI clone is technically a new fixation independent of the original recording. Therefore, the focus of the struggle in Turkish law is shifting toward moral rights.

B. Special Analysis: Moral Rights and Infringement of Voice Integrity

Moral rights express the personal bond between the author and the work; they are non-transferable and remain with the author even after the pecuniary rights are transferred. FSEK enumerates moral rights restrictively: To present the work to the public (Art. 14), to demand recognition as the author of the work (Art. 15), to prevent changes in the work (Art. 16), and to access the original of the work in the possession of another (Art. 17).

The strongest defense mechanism against AI cloning in Turkey is the Right to Prevent Changes in the Work(FSEK Art. 16). This article grants the author the authority to forbid any alteration that damages their honor and reputation or distorts the nature and characteristics of the work, even if they have given written consent.

Turkish law has the capacity to prevent infringements on artistic integrity and honor, even if technically there is no copyright infringement. AI cloning can infringe upon these moral rights through the following mechanisms:

Attribution Infringement (FSEK Art. 15): The presentation of the AI output under the name of the AI or the cloner instead of the cloned artist infringes upon the right to demand recognition as the author.
Damage to Reputation (FSEK Art. 16): The use of the artist’s cloned voice in a manner that harms their artistic understanding, honor, or professional reputation—for instance, in low-quality spam music, obscene content, or political propaganda—can be deemed an infringement of the moral integrity and reputation of the work. This broad interpretation of FSEK Art. 16 is a unique and powerful legal tool that can be used if AI deepfake technology produces content that damages the artist’s reputation.

Table 2: Ways Moral Rights can be Infringed by AI Cloning under FSEK

Moral Right (FSEK Article)	Value Protected	Possible Form of Infringement with AI Cloning
Sharing the Work with the Public (Art. 14)	Control over the time and form of publication of the work.	Early or inappropriate publication of an AI output to which the artist has not consented.
Demanding Recognition as the Author (Art. 15)	Right of attribution and correct naming.	Attributing the AI output to the cloner instead of the original author.
Preventing Changes in the Work (Art. 16)	Moral integrity, honor, and reputation of the work.	Distortion of the original vocal style in a way that damages the artist’s reputation or its use in spam content.

VII. Music Industry Response and New Business Models

A. From Legal Battle to Controlled Licensing Partnership

The music industry is adapting to the challenge posed by AI technology. Universal Music Group (UMG) decided to resolve its dispute and collaborate with the AI music platform Udio, which it had previously sued for copyright infringement. This move indicates that legal struggle has been replaced by a strategy of controlled commercial opportunity.

This partnership aims to create a “walled garden” model where fans can remix songs or experiment with vocal styles with the artist’s permission. UMG seeks to ensure that AI creations on this platform do not compete with original tracks and that artists maintain control over their work. Only songs from artists who “opt-in” will be available for AI training and remixing, and participants will benefit from royalties and potentially valuable data on how fans are interacting with their music. This “opt-in” mechanism and revenue sharing show that copyright is being redefined in the AI era to include the licensing of the artist’s artistic identity (voice, style) for use by the AI model. Licensing agreements must strictly define the context of use for this cloned voice, thereby allowing artists to secure their pecuniary and moral rights.

B. Platform Policies and Content Management

Digital music platforms are also taking “counter-armament” measures against the “escalation” brought by generative AI. In the last 12 months, during the boom of AI tools, Spotify has removed over 75 million spam-filled songs from its platform. This high number indicates that AI has become a “spam generator” capable of rapidly manipulating digital ecosystems. Spotify aims to protect artist identity by developing stricter sanctions against impersonation infringements, new spam filtering systems, and industry-standard music credits for AI use.

YouTube has taken similar measures. The platform offers its music partners the ability to request the removal of AI-generated content that mimics an artist’s unique singing or rapping voice. Furthermore, YouTube requires content creators to disclose when realistic content is made with synthetic media, and labeling may be applied to the video itself for AI content concerning sensitive topics (news, health). These platform policies create faster and more practical intra-sector regulation than legal processes, focusing on governing the malicious use of AI.

VIII. Conclusion, Legal Gaps, and Expert Recommendations

A. Summary of Key Areas of Legal Uncertainty

AI voice cloning technology creates challenges for existing legal frameworks in three key areas:

Weak Federal Copyright Protection Against Cloning: The biggest legal gap globally is the inability to categorize newly created AI fixations—which mimic the abstract quality of the voice—as direct copyright infringement. Consequently, the struggle has shifted to the field of personality rights.
Lack of TDM Regulation: In Turkish Law, the scope of commercial Text and Data Mining (TDM) activities concerning copyright is not clearly defined. This uncertainty creates a significant risk for rights holders by paving the way for the unauthorized use of copyrighted materials.
Difficulty in Distinguishing Work/Tool: As the autonomous production capability of AI increases, the distinction between whether the resulting product is a “work” in the FSEK sense or merely an “intellectual product” lacking the human labor requirement becomes challenging, threatening the authorship regime.

B. Urgent Regulatory Needs and Expert Recommendations for Turkey

In light of existing legal gaps and international regulatory trends, the following steps are needed to protect the Turkish creative industry against AI cloning risks:

Strengthening the Interpretation of Moral Rights: Explicit interpretation or supportive secondary regulations of FSEK Article 16 (Right to Prevent Changes in the Work) in the context of commercial and reputation-damaging cloning carried out by AI (e.g., low-quality or unethical content generation).
Detailing the Scope of KVKK: Mandatory explicit consent protocols regarding the use of personal data (voice) in the training of voice cloning models. The scope of consent must be detailed to include the cloning technologies and the commercial use purposes of the outputs.
Introduction of an TDM Opt-out Mechanism: Integration of an “Opt-out” (explicit objection) or express reservation right into FSEK against commercial TDM activities, parallel to the European Union model. This would grant rights holders the authority to prevent the unauthorized use of their works for AI training.
Codification of the Right of Publicity: Implementation of a specific Right of Publicity regulation within the scope of Unfair Competition Law or Civil Code to protect the commercial identities (name, image, voice) of public figures against commercial cloning, thereby solidifying protection over personality rights.

C. Strategic Conclusion

While AI voice technologies offer unique potential for the music industry’s transformation, they also introduce serious legal risks. In this era where the legal framework struggles to keep up with the pace of technology, the future of the creative industry will depend not so much on legal protection but on the success of controlled licensing models (like the UMG/Udio collaboration) and transparency protocols enforced by platforms, which guarantee artist control and royalty payments. For artists in Turkey, Moral Rights and KVKK will continue to function as the primary and strongest line of defense for protecting their identity and reputation.

Our Latest Articles

« Previous Post