Open Access Research Article
White Black Legal – International Law Journal · ISSN 2581-8503
AI-Generated Music and Copyright Law in India: A Study of the Legal Issues
Read the Full Research Paper
Access the complete open-access article in PDF format. No login is required.
Abstract
AI-GENERATED MUSIC AND COPYRIGHT LAW IN INDIA: A STUDY
OF THE LEGAL ISSUES
AUTHORED BY - DHRUVA BHARADWAJ[1]
Abstract
AI can write
melodies, orchestrate ensembles, write lyrics, record sounds and create voice
impressions at a pace and scale that defies traditional copyright
classifications. Although copyright law in India recognizes musical works,
lyrics and performances, it does not clearly define how authorship, ownership
and liability will apply to situations where a model produces music, albeit
with some human involvement. It is a doctrinal research paper which explores
the Copyright Act, 1957, important judicial decisions in India and the recent
policy consultation of the Government of India regarding generative AI. It
examines the limit of originality, the scope of the ‘person who causes’ a
computer-generated work to be created, and the protected music used in training
sets, among other issues, as well as substantial similarity in outputs, the
replication of singers' voices and the economic status of composers and
performers. The paper calls for distinguishing between protection of truly
human creative works and protection of works that are solely the product of
machines, for more transparency and viable licensing of uses for training, and
for remedies that are compatible with musical creativity and technological
innovation.
Keywords: Artificial
intelligence; musical works; sound recordings; authorship; copyright
infringement; voice cloning; royalties; India.
1. Introduction
The music
generated by generative artificial intelligence (AI) has come a long way from
computer-aided editing. It is now possible for a user to type a short
description of a genre, mood and instrumentation, and then receive a song-like
recording, with melody, lyrics and a synthetic singer.[2]
An involved creator can utilize AI to create a preliminary sketch and then
choose, rewrite, rearrange and play the material. The law needs to
differentiate these: a tool that helps a composer is not the same as a system
that produces almost everything but adds nothing in the way of a composer's
creative input.[3]
The main law
of India is the Copyright Act, 1957. The categories and licensing restrictions
were created with the assumption of an identifiable author, composer, producer
and performer. They can speak to a lot of the uses of AI music, but the Act
doesn't really create a comprehensive regime for modern generative models. It
is not just about copyrightability of ‘AI music'. Herewith comes whose
expression is protected (including authorized uses of existing works in the
production of the model), whether a specific output infringes on protected
expression, and who should exercise rights over a digitally distributed track. [4]
The paper
adopts a doctrinal approach to legal research, examining statutory provisions,
reported judgments and official policy documents up to September 2026. It poses
three interrelated questions: What kind of human contribution is needed to
enable AI-assisted song writing? When training and output creation involve the
use of existing music rights? And when does copyright and performer/personality
rights deal with synthetic voices?[5]
It posits that authorship, input licensing and output liability should not be
viewed and addressed as a single issue and promptly denied or granted rights on
every AI-generated music.
The question
is particularly topical for India's film and independent-music sectors, where a
song could be commissioned via several contracts and streamed, performed live,
or be used in short-form video and advertising. When a dispute arises over an
AI-generated track, it's not just the individual that typed the prompt who is
impacted, it can also impact lyricists, composers, labels, singers, and
distributors. The same music file can be legally created, but not legally sold
as an endorsement, or lack voice imitation but can legally be used to copy out
a protected melody.[6]
When a cure is prescribed for one harm, it is wise to classify and avoid
indiscriminately applying it to other harms.
2. Copyright
Protection of Music under Indian Law
A commercial
track isn't a one undifferentiated object. The term ‘musical work’ means music
(including its graphical notation) but not the words to be set to music.
(Section 2(p)). Possible protection as a literary work (lyrics) and as a ‘sound
recording’ under section 2(xx) of the Act. The Act also provides definitions of
‘composer’, ‘producer’ and ‘performer’ which allocate different interests to
the people that are responsible for making different contributions.[7]
All of these layers may be affected by a generative system, but no use of a
generative system will extinguish the various legal treatment of these layers.
Section
13(1) acknowledges that copyright exists in original literary works and
original musical works, in cinematograph films and in sound recordings, as
separate types of works. Section 13(4) preserves the copyright in an underlying
composition or lyrics. Section 13(3)(b) also provides that any sound recording
being made in violation of a sound recording that is a derivative of a
literary, dramatic or musical work is not eligible for copyright protection.
Therefore, a new audio file doesn't automatically lose its commercial potential
simply because its waveform was newly synthesized.
There are
other exclusive rights designated in Section 14. A musical work includes the
rights to reproduction, issuing copies, public performance and communication,
adaptation and making a recording or film in respect of the musical work, a
sound recording owner has designated rights to make further sound recordings
embodying it, supply copies and to communicate it to the public. There can be
multiple licences used in online distribution. The duty to obtain a licence to
sample a master recording does not automatically include a licence to sample
the composition, nor will a licence to use the composition automatically
include a licence to copy a particular master.
The general
rule, with statutory exceptions and agreements, is in Section 17. Assignments
are governed by sections 18 and 19, and the rights assigned, territory,
duration and consideration must be paid attention to. The changes of 2012
safeguard certain rights of authors and composers of literary works and musical
compositions used in films and sound recordings under the statutory
requirements.[8]
Labels, composers and rights owners of platforms and AI vendors, so contracts
should therefore specify what rights are covered, as well as the permitted uses
for machine learning, not blanket ‘all media’ clauses which may have assumed
that all issues have been addressed.
3. Authorship, Originality and Ownership of
AI-Generated Music
The composer is identified in section 2(d)(ii) as the
author of a musical work and the producer is identified in section 2(d)(v) as
the author of a sound recording. Most importantly, section 2(d)(vi) calls the
‘person who causes the work to be created’ the author of a computer-generated
literary, dramatic, musical or artistic work. The latter phrase makes Indian
law unique but it doesn't explicitly say an AI model is a legal author or that
anyone hitting ‘generate’ owns all the resulting melodies. It is context
sensitive who is exactly the person that causes creation, even within the scope
of a legal framework.
Consider three situations. The first is the basic
human melody and lyrics, with AI orchestration; the original human composition
might be recognizable. Secondly, the producer provides precise musical guidance
and repeatedly alternates between outputs, make significant changes to the
expressive elements, the issue is what parts of the expressive elements have
been changed based upon the producer's original direction. Thirdly, a user asks
for ‘a happy pop song', and the first unedited response is posted. If they were
all considered equally created by the user, it would be difficult to
distinguish differences in the contribution to the creation and would allow for
broad monopolies on material generated mostly through automation.
This is where the originality requirement comes into
play. The Supreme Court in the case of Eastern Book Company v. D.B. Modak[9]
held that protection under the Copyright Act could not be granted on the basis
of labour alone and a certain creative effort in the writing of the work was
required. Although the judgment did not address the question of AI authorship
of music, it does provide some guidance on how to think about originality, and
its reasoning indicates that the act of choosing to use AI tools – even if a
subscription fee was paid – is insufficient to afford a protectable human
musical expression. On the other hand, the lyrics written by a human, melodies
transformed in some way, new arrangements, and/or composition or selection of
new music can receive their own evaluation.
A new recording may be made by the producer of the
newly recorded production and the owner of those rights may assert the rights
that apply to the sound-recording layer with respect to that layer. This is not
the end of the discussion of ownership of an underlying composition created by
AI. Nor does the wording of a platform's terms of service definitively
establish whether there is a right in the first place in the Indian statute: a
contract can award rights or grant rights to use a right that is present or has
been promised; it cannot create a right out of thin air. The real solution is
to record human contributions, prompts, successive versions, performance
credits and contractual chains of title.
A practical distinction is between a claimant's
contribution to a specific contribution, and ownership of the completed
release. The bridge of a song can be written by a collaborator who does not
write the chorus, or the song can be recorded by a collaborator without being
the owner of the words. When AI is a part of a collaborative process, the
credits must clarify which elements of the collaboration are being claimed as
AI-generated and agreements should include a discussion of whether rights are
being licensed for the final recording. This is a more defensible approach than
claiming an undifferentiated ownership of a whole AI-assisted track prior to
locating the legally protected contributions.
4. Training on Copyrighted Music: Licensing and
Exceptions
The statistical relationships that are learnt in
generative music systems are usually established from a large set of
recordings, compositions, lyrics and related metadata. In addition to any
specific song produced by training, there are input-side questions arising from
training which cannot be answered without examining the song: has a protected
work been copied or used in other ways that would involve the exclusive rights
of the copyright holder; and, if so, is it licensed or protected by a statutory
exception? Also, the different datasets have various layers of legal
protection, such as underlying compositions and master recordings. The use of a
public stream or the acquisition of a song is not sufficient to create a
licence for model development.
Section 51 covers infringement if someone does
something which is normally protected by copyright that they don't have
permission to do and not under an exception. There are purpose exceptions in
Section 52(1)(a) such as private or personal use (including research),
criticism or review and reporting current events. It isn't a waiver of all
commercial training just because training can be called ‘research’. Section 52
also covers specific transient or incidental electronic use in specific
situations, which may or may not be met by the use of a particular AI-training
pipeline, depending on the nature of its use and the conditions set out in the
statute.[10]
When copyrighted sounds are downloaded, normalised,
segmented and/or stored in order to be used repeatedly for training, arguments
about temporary technical copying should be taken seriously rather than taken
for granted. Transparency is also important: If there are no records of where a
composer got the work and the model cannot be identified, the composer may not
be able to tell; if there are no records of where a developer got the work, the
developer may have a difficult time showing licensing or relying on an
exception. Copyright compliance is thus also an evidence and data-governance
challenge and not just an abstract question about whether a trained model
contains a literal copy.
The Department for Promotion of Industry and Internal
Trade (DPIIT) released a working paper on generative AI and copyright for
stakeholders' consultation in December 2025, which is Part I of the working
paper. It discussed various models such as exemptions and opt-outs, voluntary
and collective licensing as well as a ‘One Nation, One License, One Payment’
approach for training uses. That is a proposal for a policy, not a statute. The
publication of the document represents an official acknowledgment of the creator-remuneration
and innovation clash, but should not be referred to as if it has amended
sections 14, 51 or 52. [11]
In the case of music, licensing design needs to
consider the fact that compositions, lyrics, and recordings are subject to
several different types of ownership interests.In the case of music, licensing
design must recognize the fact that there are several different types of
ownership interests for compositions, lyrics, and recordings. This could
circumvent composers or other statutory beneficiaries as the model would be
conveniently administered based on the ownership of masters. Frequently,
however, a bespoke negotiation for each track can be costly for smaller
domestic developers. All future systems should provide use reporting, include
explicit distribution mechanisms, and have realistic use and dispute resolution
procedures without equating all works with all uses.
Consent should also differentiate between model
training and sampling and generating a close imitation. The artist may agree to
a model being analysed in a more general sense to make it better, but reject
the use of the model to produce a commercial product featuring a promotion of
the artist's ability to replicate their voice. On the other hand, a label might
be able to license a recording it owns but not be able to license an underlying
piece of music or a singer's voice. There is a need for clarity in granularity
because if everything is lumped together as ‘AI use’, it becomes challenging to
contract and audit the use of AI in the future.
5. Generated Outputs: Substantial Similarity, Sampling
and Synthetic Voices
The question
that is raised at the output is whether a track is protected expression of an
earlier one. In R.G. Anand v. Deluxe Films[12],
the Supreme Court clarified the distinction between ideas and themes and
protectable expression, and noted that substantial copying and material copying
are distinct ideas. The music and AI cases were useful, but not this one: The
use of a wide genre, a familiar chord progression or overall mood is not
necessarily the same as the copying of a substantial and original melody or a
memorable lyric section. Similarity is to be judged in reference to the actual
work protected and the nature of the taking.
Direct audio
sampling needs to be given special consideration. An AI-assisted track may
include a recognizable portion of a master track, in which case there is a
sound-recording right that can be exercised without regard to an underlying
tune being original enough. A system can also duplicate a signature, or a
particular phrase with a melody, without copying the original master's sound.
Not all resemblance is infringement and an allegation must be carefully
compared and compared with evidence of provenance and assessment of the
substantial protected features, not just a numerical similarity score provided
by software.
Under
section 31C of Indian law, cover versions are treated separately, provided that
the version is transmitted on a non-exclusive basis, and under section 38, 38A
and 38B of the law, a number of rights of performers are recognised. A track
that is labelled as a ‘cover’ and/or has a vocalist with an AI does not
automatically mean that no composition licence, recording rights, performance
rights or statutory regime apply. Royalties also vary according to
exploitation; contractual chain and statutory principles, not a general rule,
govern who is entitled to receive the revenue. Royalties also are dependent
upon the type of exploitation; not a universal rule that an AI user must
receive all revenues, but a contractual chain and statutory principles are
considered. [13]
Voice
cloning poses another challenge. A synthetic voice can sound like a familiar
popular singer, but not be the same as any of the segments of the sound
recording. Training and outputs also might engage copyright individually, but
copyright may not be an adequate solution. In Arijit Singh v. Codible Ventures
LLP[14],
the Bombay High Court in July 2024 issued interim directions in relation to the
unauthorised use of a singer's voice and other personality features in an
AI-generated video. It is an important decision because the "evidential
facts" presented give some commercial and reputational weight to synthetic
performance, but it is an interim ruling, not a blanket prohibition against all
stylistic parallels. This is a noteworthy decision on its merits because it
gives some commercial and reputational weight to the fact of synthetic
performance on the presented facts, but it is an interim decision, not a
general prohibition on all stylistic parallels.
In the case
of Anil Kapoor v. Simply Life India[15],
the Delhi High Court has shown itself ready to enjoin some forms of
unauthorised commercial use of a person's identifiable voice and persona using
technological solutions. Where their individual statutory conditions are met,
an author's special rights under section 57, and performer moral rights under
section 38B, may also be applicable. The doctrinal categories are not supposed
to be combined: Mimicking a voice, copying a protected recording, passing off
an endorsement and distorting an actual performance can be different causes of
action and evidentiary requirements.
6.
Liability, Remedies and the Emerging Policy Framework
The
ownership of an AI song can be shared by the creators of the AI model, the
creators of the dataset used to train the model, the creators of reference
songs, the users who interacted with the AI, and the producers and platforms
that generated the AI song. The current copyright principles demand
determination of what is protected, which restricted act and what specific
actions of a party. Reliance on any single developer to be responsible for any
similarity created by any user is too much to ask, as is expecting any user to
be protected because an audio model is created. The enquiry into the evidence
should follow the facts of the copying, the design of the system, the giving of
notice, the authorisation for the copying and the control of it, where these
issues are relevant to the pleaded legal claim.
Section 55
of the Copyright Act gives civil remedies for copyright infringement to include
injunctions, damages, accounts and of course, separate enforcement of divisible
rights under section 56 of the Copyright Act. In addition, section 57 covers an
author's specified moral interests, while section 38B does a similar job with
respect to the performers. In practice, those wishing to seek urgent relief
against the use of a synthetic song would need to state what song, composition,
lyric, actual performance or persona is being misused and what specific acts
are complained of. Orders must be specific enough to remedy actual
infringement, rather than stifle legitimate non-infringing musical expression.[16]
Conflict
could be minimised with collective licensing and improved metadata. A viable
system could separate training from output exploitation, indicate to the
relevant owners of the data and to the statutory beneficiaries of the datasets
what is involved, but not reveal all of the "edens," and enable a
dispute over the attribution of the datasets. Developers might maintain records
of provenance, provide useful reporting and never promote characteristics that
deceptively suggest a singer's consent. Labels and musicians would be able to
discuss the use of artificial voices and training agreements that would be
explicit, rather than leaving it all up to the old recording contracts.
The DPIIT
working paper should be used as a basis for discussion and dialogue and is not
intended to replace legislation or binding judicial interpretation. Due to the
border-crossing nature of AI-generated music, reform must also be given
attention to the enforcement and different approaches to national authorship.
India can't import a foreign model wholesale: its definition of
computer-generated-work, layered music rights, protections for performers and
provisions for royalties require a coherent response at home. Any new regime
should be explicit as to whether it is about works being ingested or rights
being outputted, or about both.
Provenance
information is accessible and can be used for enforcement. Knowing who has made
a model and what version of it is being used, from where the reference audio
was uploaded, and successive human revisions do not necessarily have to be
disclosed to the public, nor do the weights have to be all made public, to be
proprietary. With such a supervised process there is a possibility to allow a
court to review relevant evidence without compromising truly confidential
evidence. Platforms should also include notices that differentiate between
claims of master-recording copying, melodic similarity and impersonation, where
the proof and legal grounds for each complaint are different.
7. Conclusion and Suggestions
AI-generated
music poses several legal questions, not one. A song can contain protectable
human lyrics, a contested AI-generated melody, a newly produced recording and
an imitation of a recognised performer’s voice. Each attracts a different
statutory enquiry. Section 2(d)(vi) provides a possible route to identifying
the person behind a computer-generated work, but it does not conclusively
allocate authorship of every autonomous output. Originality and actual human
creative contribution remain important, as do the independent rights in source
material used for training and in any expression reproduced in the final track.
A balanced
Indian response should adopt four principles. First, protect identifiable human
composition, selection, arrangement and performance without granting automatic
monopolies over all material produced by a prompt. Secondly, require clear,
workable authorisation or a carefully legislated alternative for protected
training uses, with remuneration that reaches the appropriate music
right-holders. Thirdly, provide effective, tailored procedures against
substantial copying, misleading impersonation and unauthorised exploitation of
distinctive voices. Fourthly, improve evidence preservation, model transparency
and contractual clarity so that disputes can be resolved on facts rather than
speculation.
The
distinction between innovation and appropriation should guide reform. Human
musicians should not lose their legal interests merely because their work
becomes useful training material; equally, copyright should not be extended so
broadly that it captures genres, techniques or ordinary musical ideas. The
Copyright Act’s existing framework offers substantial guidance, but legislative
clarity on training, attribution and the status of largely autonomous outputs
would better protect both creative labour and responsible development of Indian
music technology.
BIBLIOGRAPHY
Statutes and Legislation
The Copyright Act, 1957 (Act 14 of 1957).
The Copyright (Amendment) Act, 2012 (Act 27 of 2012).
Cases
Anil Kapoor v. Simply Life India & Ors., 2023 SCC OnLine Del 6914.
Arijit Singh v. Codible Ventures LLP, 2024 SCC OnLine Bom 2445.
Eastern Book Company v. D.B. Modak, (2008) 1 SCC 1.
R.G. Anand v. Deluxe Films, (1978) 4 SCC 118.
Research Papers
Andrea Agostinelli et al., “MusicLM: Generating
Music From Text” (2023), available at:
[https://arxiv.org/abs/2301.11325](https://arxiv.org/abs/2301.11325).
Prafulla Dhariwal et al., “Jukebox: A Generative
Model for Music” (2020), available at:
[https://arxiv.org/abs/2005.00341](https://arxiv.org/abs/2005.00341).
Government Reports and Policy Documents
Department for Promotion of Industry and Internal
Trade, Government of India, Working Paper on Generative AI and Copyright, Part
I: One Nation One License One Payment—Balancing AI Innovation and Copyright
(Dec. 2025), available at: [https://www.dpiit.gov.in/static/uploads/2025/12/ff266bbeed10c48e3479c941484f3525.pdf]
(https://www.dpiit.gov.in/static/uploads/2025/12/ff266bbeed10c48e3479c941484f3525.pdf).
[2] Andrea
Agostinelli et al., “Music LM: Generating Music From Text” (2023), available
at: https://arxiv.org/abs/2301.11325; Prafulla Dhariwal et al., “Jukebox: A
Generative Model for Music” (2020), available at:
https://arxiv.org/abs/2005.00341.
[5] The
Copyright Act, 1957, ss. 38A and 38B; Anil Kapoor v. Simply Life India,
2023 SCC OnLine Del 6914.
[6] The
Copyright Act, 1957, ss. 13(4), 14(a), 14(e) and 51; Anil Kapoor v. Simply
Life India, 2023 SCC OnLine Del 6914.
[8] The
Copyright (Amendment) Act, 2012 (Act 27 of 2012); The Copyright Act, 1957, ss.
17, 18 and 19.
[11] Department
for Promotion of Industry and Internal Trade, Government of India, Working
Paper on Generative AI and Copyright, Part I: One Nation One License One
Payment—Balancing AI Innovation and Copyright (Dec. 2025), available at:
[https://www.dpiit.gov.in/static/uploads/2025/12/ff266bbeed10c48e3479c941484f3525.pdf](https://www.dpiit.gov.in/static/uploads/2025/12/ff266bbeed10c48e3479c941484f3525.pdf).
How to Cite This Article
DHRUVA BHARADWAJ, AI-Generated Music and Copyright Law in India: A Study of the Legal Issues., White Black Legal – International Law Journal, ISSN: 2581-8503, Vol. Volume 4, Issue VOLUME 4 ISSUE 1, September 2026, pp. 1045-1057. Available at: https://www.whiteblacklegal.co.in/details/ai-generated-music-and-copyright-law-in-india-a-study-of-the-legal-issues
Author & Publication Record
Authors: DHRUVA BHARADWAJ
Registration ID: 107120 | Published Paper ID: WBL7120
Year: Sep- 2026 | Volume: 4 | Issue: 1
Approved ISSN: 2581-8503 | Country: Delhi, India
Page No.: 1045-1057
Full Text Preview
Open in New Tab
Copied