Can You Own What You Can Sell? AI-Generated Music and Copyright under Serbian Law
A Song Created in Seconds
Imagine creating a song in seconds, uploading it to a streaming platform and earning money from it, without ever composing a melody yourself. A person in Serbia uses an AI music-generation platform, enters several instructions, and receives a complete song. After purchasing a subscription, the user downloads the track and distributes it commercially. The song begins generating revenue.
But does the ability to monetise it mean that the user owns its copyright? Can the song qualify for copyright protection, and what exactly has the subscription secured?
These questions are particularly relevant to platforms such as Suno, whose subscription guidance expressly distinguishes commercial use rights from copyright protection. A user may be permitted to earn money from a generated song without being guaranteed copyright ownership. This distinction reveals a broader legal uncertainty. Under Serbian law, the right to commercially exploit music, the existence of copyright protection and the identification of its author are separate questions. Understanding their relationship is essential for users who increasingly rely on generative AI in creative industries.
Can an AI-Generated Song Have an Author?
Under Article 2 of the Serbian Copyright and Related Rights Act, a copyright-protected work must constitute an original intellectual creation expressed in a particular form. The provision expressly includes musical works, with or without lyrics. Article 9 defines the author as the natural person who created the work. These provisions reflect the human-centred foundations of Serbian copyright law. The challenge posed by generative AI is therefore not whether music belongs to a category of protectable works, but whether the particular composition embodies an original human creative contribution.
Article 6 further excludes general ideas, methods and instructions as such from copyright protection. This distinction becomes particularly significant when users generate music through written prompts. An instruction requesting a melancholic jazz composition may communicate an idea or desired style without necessarily determining the protected expression of the final song. Although Serbia is not an EU Member State, the case law of the Court of Justice of the European Union offers a useful comparative perspective.
In Painer (C-145/10, paragraphs 87–94), the Court connected originality with the author’s free and creative choices. Examining portrait photography, it recognised that such choices may occur during preparation, production and subsequent processing. The use of technological equipment does not, by itself, exclude human authorship. Similarly, in Cofemel (C-683/17, paragraphs 29–35), the Court reaffirmed that a protected work must be an original intellectual creation expressed through identifiable subject matter. Its reasoning also makes clear that aesthetic appeal cannot substitute for originality. Applied by analogy to AI-generated music, these principles suggest that the decisive question concerns the relationship between the user’s creative choices and the musical expression produced. A commercially successful AI-generated song is not necessarily protected merely because it sounds professional or attracts listeners. Conversely, the involvement of AI should not automatically deprive a genuinely human-authored musical creation of protection. The relevant issue is whether, and to what extent, human originality can be identified in the final result.
When AI Is a Tool, Not a Creator
The distinction between AI as a creative tool and AI as an autonomous generator is central to determining authorship. Consider three situations. A user may enter a simple prompt and allow the system to generate an entire song. Alternatively, the user may provide original lyrics while relying on AI to generate the accompanying music. In a third scenario, the user may progressively develop the composition through carefully designed and iterative prompts specifying musical elements such as melody, harmonic progression, rhythm or arrangement, as well as through subsequent creative modifications. Such instructions may reflect the user’s creative choices and influence the overall musical expression of the final composition.
These situations involve different degrees and forms of human participation, and should not necessarily produce identical legal outcomes. In Understanding Authorship in Artificial Intelligence-Assisted Works, Johannes Fritz proposes an assessment that considers who uses the system, the nature of the technology, the user’s creative judgment and the degree of control over the execution of those decisions. His approach draws attention to whether human choices are actually reflected in the resulting work. Christophe Geiger similarly emphasises that human creative intervention may occur at different stages of production. Generative AI can serve as an instrument of human creativity, provided that the relevant requirements of originality are fulfilled. A recent judgment of the Munich Local Court (Amtsgericht München), a German first-instance court, illustrates the difficulty of applying these principles to AI-generated content. In its judgment of 13 February 2026 (142 C 9786/25), the court examined AI-generated logos produced through detailed and iterative prompts. Although not binding under Serbian law, the judgment offers a useful comparative perspective on the assessment of human creative contribution. Importantly, however, the court did not categorically exclude copyright protection based on prompting. It recognised that creative human intervention during the generation process or through subsequent modifications may be relevant.
This distinction deserves particular attention in the context of music. A carefully designed prompt may involve more than specifying a genre or general atmosphere. It may express decisions concerning the structure, progression, arrangement or other identifiable musical elements of the intended composition. Nevertheless, the mere complexity or length of a prompt should not automatically establish authorship. What matters is whether the user’s creative decisions meaningfully shape the final expression rather than leaving its essential creative features to the system.
In our view, Serbian law should allow recognition of human authorship in AI-assisted music where an identifiable original contribution is established, including potentially through carefully developed and iterative prompting. Such recognition should depend on the circumstances of each case, not on a presumption that every AI-generated output belongs to its user.
A Commercial Licence Is Not Copyright
Even where an AI-generated song does not qualify for copyright protection, a user may still be contractually permitted to exploit it commercially. This is where the distinction between contractual rights and copyright becomes especially important. Under Suno’s Terms of Service, effective from 3 September 2026, Pro and Premier subscribers may receive an assignment of Suno’s rights, title and interest in certain generated outputs, subject to the applicable contractual conditions. The wording is significant. Suno assigns rights that it holds in the relevant output, while expressly declining to guarantee that copyright will arise in that output. This creates two separate legal questions: what rights does Suno possess and agree to assign, and whether the generated material satisfies the legal requirements for copyright protection in the first place.
A contractual assignment may allocate existing rights, but it cannot independently establish copyright in material that does not qualify for protection under the applicable law. Suno also regulates commercial exploitation separately. Its terms permit commercial use of qualifying outputs obtained through authorised download channels and impose restrictions on unauthorised downloading and the concealment of certain identifying information attached to generated content. Particularly noteworthy is Suno’s express characterisation of these restrictions as contractual obligations between the platform and its users, applicable regardless of whether copyright or another proprietary right exists in the output. Furthermore, the terms provide that qualifying assignments and commercial-use rights associated with permitted downloads generally survive the expiry or cancellation of a subscription. The platform’s guidance for paying subscribers illustrates the commercial opportunities involved. Users may monetise eligible songs through streaming services, independent sales or incorporation into films, television productions and video games. However, the availability of these opportunities does not establish that the subscriber is the author.
Serbian copyright law also distinguishes authorship from the subsequent acquisition of economic rights. Under Article 9, a copyright holder may be someone other than the original author. Articles 61–63 regulate the assignment of economic rights, including exclusive and non-exclusive arrangements. Yet an assignment presupposes an underlying right capable of being assigned. Commercial permission granted by a platform should therefore not be confused with exclusive copyright protection enforceable against third parties. Nor should the terms of an international platform be treated as determining, by themselves, the copyright status of a composition under Serbian law.
The essential distinction is that a user may acquire contractual permission to monetise an AI-generated song without necessarily acquiring copyright ownership or the ability to prohibit its use by others.
Who Can Profit — and Who Can Claim Rights?
The practical consequences of this distinction become clearer when considering the position of third parties. Suppose a Serbian subscriber generates a song through Suno, downloads it in accordance with the platform’s rules and begins earning revenue from its distribution. Another user subsequently obtains an identical or substantially similar composition and publishes it independently.
Can the first subscriber prevent that publication? The answer cannot be based solely on payment, priority of generation or the existence of commercial revenue. Under Articles 19 and 20 of the Serbian Copyright and Related Rights Act, copyright provides economic rights relating to protected works, including the exclusive right to authorise or prohibit reproduction. Article 62 additionally distinguishes exclusive from non-exclusive assignments of economic rights. Where copyright protection exists, the identity of the relevant right holder and the scope of any assignment become essential. Where the musical composition does not qualify for copyright protection, the first user’s contractual permission to monetise it does not, by itself, create an exclusive copyright claim against another person. This distinction is particularly relevant because Suno acknowledges that its system may generate identical or similar outputs for different users. Nevertheless, the absence of copyright in an entirely AI-generated composition does not necessarily exclude protection for individual human-authored contributions.
Original lyrics written by the user, independently composed musical passages or sufficiently creative arrangements may qualify for protection where they satisfy the statutory requirements. Such protection should be evaluated by reference to the particular elements created by the human author, rather than automatically extending to the entire AI-generated result. A further qualification concerns related rights. Serbian law separately protects certain rights of phonogram producers under Articles 123–126. Under Article 125, a phonogram producer is a natural or legal person who organises and finances the production of a sound recording and bears responsibility for its first fixation. These rights concern the sound recording itself and are distinct from copyright in the underlying musical composition. However, whether a particular AI-generated recording gives rise to such rights, and who qualifies as its producer, requires an independent legal assessment. Commercial exploitation may also be affected by the rights of third parties and the distribution rules of external platforms. Suno’s contractual permission is therefore not an unconditional guarantee that every use of a generated song will be legally unobjectionable. For users, the resulting distinction is substantial. A subscription may create an opportunity to distribute and monetise music, but it does not necessarily establish exclusivity, authorship or freedom from competing legal claims. The possibility of profiting from music and the legal ability to prevent others from exploiting it are different forms of entitlement.
What Should Serbia Clarify?
Serbian copyright law already contains the essential foundations for protecting human creativity: originality, expression in a particular form and authorship attributed to a natural person. The central challenge is determining how these established concepts should apply when the creative process involves generative AI. In our view, Serbia should avoid two equally rigid approaches: automatically recognising the user of an AI system as the author of its outputs, and automatically excluding copyright protection whenever generative AI participates in the creative process. Instead, relevant institutions should consider developing practical interpretative guidance for assessing human creative contributions in AI-assisted works. Such guidance could address the specificity and creative substance of prompts, the extent to which a user controls identifiable expressive elements, the relevance of iterative instructions, original input material and subsequent creative modifications, and the evidence needed to demonstrate the user’s contribution. For this reason, maintaining a record of the creative process, from initial prompts and successive AI-generated versions to the user’s own refinements, could help establish how individual creative choices shaped the final musical work. These factors should guide an individual assessment rather than operate as rigid numerical thresholds. Neither a minimum number of prompts nor a fixed percentage of human involvement would necessarily capture the originality of a musical composition. Clearer guidance should also distinguish contractual permission to commercialise AI-generated content from the existence and scope of exclusive rights under copyright law. Users should be able to understand what a subscription grants, what it does not guarantee and which legal questions remain unresolved. The comparative reasoning in Painer, Cofemel and the Munich judgment, together with the approaches developed by Fritz and Geiger, provides useful material for such clarification. The objective should be to bridge the gap between traditional concepts of authorship and contemporary AI-assisted creative practices without weakening the central role of human creativity. Ultimately, the ability to sell an AI-generated song does not necessarily mean owning its copyright. A clearer legal framework should help users understand both the possibilities and the limits of that distinction.