Name of author: Kabir Thakur (II Year, Chanakya National Law University, Patna)
Keywords: Artificial Intelligence (AI), Music, Intellectual Property (IP), Authorship, Ownership Rights
Introduction
In the contemporary scenario, rampant use of Artificial Intelligence (AI) and technology is prevalent. Generative AI has recently boomed as a major entity in the same. It refers to any machine learning model that is trained on datasets and is capable of dynamically generating outputs by itself. With the emergence of generative AI, there has evolved a new sub-category in each field of content creation: AI-Generated Content. In terms of music, it makes no exception. AI-Generated music is a common illustration of what the generative AI can produce. It can produce a song written by X artist in the voice of another artist Y. Peter Jackson’ AI helped Paul McCartney extract John Lennon’s vocals and use it to create what is known as the last song of The Beatles – Now and Then. There are instances where artists create songs using AI to sing like famous singers. But this also leads to a big fundamental question: Are the current laws competent enough to govern and safeguard AI-generated music? What would be the authorship and copyright status of the work generated by AI? And what if the AI voice is used by an artist for his original content. The D.O.C. takes the assistance of AI to generate original vocals having his voice irreversibly altered due to a car accident in 1989. Does he not have the copyright over his work?
Understanding AI in Music Composition
How AI Composes Music?
AI uses various tools to generate new content music. It uses datasets available on the internet having symbolic representations, e.g., MIDI files or raw audio. It extracts and learns rhythms, patterns and features via machine learning, and produces its own musical composition. Machine Learning Models are highly useful in the realm of music with being adaptive as required amongst different cultural and individual needs. Deep learning neural networks are even better as they can manage complex, high-dimensional musical data and classify & extract them efficiently. Also, it uses Generative Adversarial Networks (GANs) to not be a mere re-user of gathered data, rather be able to generate original content. GAN aids in this process as making it capable of imitating the functions of a human brain and hence have its own musical sense. Prominent AI music tools like AIVA, Amper Music and Google’s Magenta have been quite successful in recent times. AIVA, a virtual composer recognized by the SACEM, contains a plethora of orchestral music and hence has immensely deep knowledge in terms of classical music. Meanwhile, Amper provides its users with the ease to compose music as per their needs and it helps in customizing the music to fit “the style, length and structure of their projects”. MAgentA recognizes the mood and generates adaptive background music in sync with the environment via the implementation of algorithms using empirical evaluation.[i]
Human Input v. Machine Autonomy
There are various arguments that in case if AI generated music is almost at par with the IP-eligible human-generated work, then there shall be no second thought about providing copyrights to the AI generated outputs. But, when seen from a broader perspective, IP eligibility is not just based on the result output but also on the process of its making, e.g., also considering the artist’s thought process and the flow of creation.
Intellectual Property Frameworks: A Primer
Moral Rights & Performer’s Rights
Article 6bis(1) of the Berne Convention protect the moral rights of an artist of their works to claim authorship of their works even if they transfer its economic rights as this type of right is inherently personal to the human creators, therefor safeguarding them from any form of AI misuse.[ii] Whereas Articles 7-9 of the Rome Convention protect performers’ rights of human artists, leaving AI-synthesized or virtual performers in doubt.[iii]
Authorship & Originality
Different jurisdictions though vary in their opinions on the stance of authorship regulation and IP protection, presuppose human interference and consider AI more as a tool rather than a whole-soul content creator. US laws believe in the essentiality of human authorship, while the EU also leaves discretion of copyright to national laws but calls for transparency in cases where the content is AI-generated.[iv] UK on the other hand allows copyright to AI-generated music but the authorship is allotted to its human “supervisor”.[v] Originality also decided by the creative choices rather than mere labour in producing it.[vi]
Collision Point: Challenges of Applying IP Law to AI Music
Authorship Dilemma
While section 9(3) of UK’s CDPA 1988 leaves some room for AI autonomous creativity, it still designates the human as “author” of its work. US and EU are even more rigid, outrightly rejecting any idea of AI-generated work’s IP protection. This raises the question whether any product resulting completely from AI would ever be considered an original piece of work and get IP protection.
Ownership Battles
Developers and artists assert their rights over the training data and algorithms of generative AI’s works. Users of these AI models support their cause by arguing that their prompts and modifications constitute as “arrangements” and hence they shall be the owner of resulting output work. Various platforms like Boomy and Mubert assert ownership over the generated music via terms of service and end user service agreements. Due to the non-clarity in this domain from the legislative end, litigation is the ultimate deciding tool.
Originality in the Algorithmic Age
Another flaw that is spotted in AI-generated content is that it lacks the ‘modicum of creativity’ demanded by the Feist standard (US 1991). It is based on data patterns as it generates the output by extracting inputs and patterns from datasets. The EU’s Pelham ruling (2019)[vii] lays down the idea that while even short human-authored sequences may be granted IP protection, AI outputs would fall short of “intellectual creation” required for its eligibility.[viii] Also, a similar issue is faced by India’s Copyright Act 1957 which has till date no reference of AI. Hence, this highlights the issue that whether AI’s recombination and paraphrasing of datasets match up to the parameters of originality or not.
Jurisdictional Splits
In US, the laws advocate for strict human authorship, mandating it as an essential for IP protection eligibility. While EU offers some width, defining AI as a tool though still requiring human input, UK on the hindsight protects AI-generated works but denies authorship rights to them, giving them to the human “arranger” instead. India though, has no existing regulatory framework for AI works, the National Strategy for Artificial Intelligence, 2018 being the only official legislation.
Possible Legal Approaches and Global Trends
Overview of Proposed Framework
The legal framework of AI-generated music can be improved by implementation of a few possible approaches. Maintaining the traditional idea of human-centricity in copyright and giving no authorship to AI, in alignment with the stance of EU and US, can be an effective method. Also, sui generis rights may be provided to AI outputs and categorization of them is needed. This would also work in consonance with how UK believes in AI as a tool while giving it copyrights on certain outputs, though not providing authorship to AI, leading it to be vested in the human “arranger”. The public domain approach treating AI outputs as inherently uncopyrightable as was stated in Thaler v. Perlmutter (2025)[ix] highlighting that it is generated because of recombination of large datasets in the public domain is also a positive idea. Though this promotes open access, investments in AI tools might suffer.
International Policy Discourse
The EU AI Act, 2024 demands transparency in AI-generated works but does little to redefine authorship. On the other hand, Article 4 of the 2019 EU Copyright Directive establishes an exception for cases where copyrighted works are reproduced and extracted for text and data mining for research purposes. It safeguards human creativity while promoting the use of copyrighted work for research purposes to an extent. UKIPO and USPTO have also rejected the idea of AI authorship, though they have explored licensing models for AI training data.
Conclusion
There is an urgency to redefine and modernize IP laws in context of AI-generated works. There are significant gaps in the copyright and authorship frameworks. Legislators need to allow the adaptation of innovative AI approaches and incentives while simultaneously safeguarding the rights of human artists. The EU’s stance of mandatory transparency about AI use in musical compositions shall be implemented alongside other regulations as a part of attribution mechanisms. When being trained by scrap copyrighted data, there shall be a revenue-sharing model, e.g., collective licensing pools. A standard-uniform framework alongside global harmonization is crucial to prevent forum-shopping and rights fragmentation.
[i] Eduardo Reck Miranda,‘Readings in Music and Artificial Intelligence’ in Eduardo Reck Miranda (ed), Readings in Music and Artificial Intelligence (Springer 2000).
[ii] Berne Convention for the Protection of Literary and Artistic Works (adopted 9 September 1886, last revised Paris Act 24 July 1971) art 6bis.
[iii] International Convention for the Protection of Performers, Producers of Phonograms and Broadcasting Organisations (adopted 26 October 1961, entered into force 18 May 1964) arts 7-9.
[iv] Regulation (EU) 2024/1689 of the European Parliamenet and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) art 52(3).
[v] Copyright, Designs and Patents Act 1988, s 9(3).
[vi] Feist Publications Inc v Rural Telephone Service Co 499 US 340 (1991).
[vii] Case C-476/17 Pelham GmbH v. Hütter EU:C:2019:624.
[viii] Copyright Act 1957 (India).
[ix] Thaler v Perlmutter No 23-5233 (DC Cir, 18 March 2025).