Few questions in contemporary intellectual property law have generated as much urgency, as much disagreement and as much genuine legal uncertainty as the question of who if anyone owns the copyright in a work generated by artificial intelligence. The question is not merely academic. Generative AI systems now produce novels, compose music, paint images, write legal documents, generate software code and create visual art at a scale and speed that no human creator could match. These outputs have commercial value. They are being sold, licensed, published and deployed in products that reach millions of users. The legal framework that determines whether these outputs attract copyright protection and if so in whom that protection vests, will shape the economics of the creative industry for decades to come.
The difficulty of the question lies in its collision with a foundational assumption that the entire architecture of copyright law has taken for granted since its inception: that creative works are made by human beings. Copyright law was designed to protect and incentivise human creativity. Its central concepts authorship, originality, expression, moral rights are human concepts, premised on the existence of a thinking, feeling, choosing author whose creative personality is expressed in the work. When the work is generated not by a human author but by a statistical model trained on billions of examples, the foundational assumption fails and the framework built upon it begins to crack.
India’s Copyright Act, 1957 was not written with generative artificial intelligence in mind. Its provisions governing authorship, originality and first ownership were drafted for a world of human creators, supplemented only by a narrow provision for computer-generated works inserted long before the current generation of AI systems existed. The task of applying this framework to AI-generated outputs requires both careful statutory interpretation and, ultimately, legislative reform that the Act has not yet received.
This article examines the copyright question for AI-generated works comprehensively moving through the nature of generative AI and why it challenges copyright doctrine, the Indian statutory framework and its specific provision for computer-generated works, the international landscape and comparative approaches, the training data question, the output infringement question, the moral rights dimension, the commercial implications for the Indian technology and creative industries and the case for legislative intervention.
The Nature of Generative AI and Why It Challenges Copyright
To understand why AI-generated works present such difficulty for copyright law, it is necessary to understand, at least in broad terms, how generative AI systems work and what makes their outputs distinctive from other computer-generated content.
Generative AI systems the large language models that produce text, the diffusion models that generate images, the transformer architectures that compose music are trained on massive datasets of existing human-created works. During training, the system processes enormous quantities of text, images, audio or code and learns the statistical patterns, relationships, structures and associations that characterise the training data. It does not store copies of specific works in a retrievable database it encodes the patterns of the training data in billions of numerical parameters. When prompted to generate an output, the system uses these learned patterns to produce a new sequence a text, an image, a piece of music that statistically resembles the training data in its structure and characteristics without, in the ordinary case, reproducing any specific training example verbatim.
The output of a well-functioning generative AI system is therefore genuinely new in the sense that it is not a copy of any existing work. But it is also genuinely derivative in the sense that every aspect of its structure, style and content reflects patterns learned from existing human works. It is, in a meaningful sense, a statistical distillation of human creativity expressing the regularities of human creative production without the specific expressive choices that any individual human author would make.
This creates the first problem for copyright law: the output is new, but it was not created by a human author exercising creative judgment in the sense that copyright law has always understood. The AI system does not have intentions, does not make aesthetic choices in the human sense, does not express a personality or perspective and cannot hold rights. The human who interacted with the system who provided the prompt, adjusted the parameters, selected from multiple outputs contributed to the process but may or may not have contributed the kind of creative expression that copyright law recognises as authorship.
The second problem is the training data. The statistical patterns that enable the AI system to generate creative outputs were learned from copyrighted works. Whether the training process the ingestion and processing of millions of copyrighted texts, images and musical compositions constitutes infringement of the reproduction rights in those works is a question of enormous commercial consequence. If training on copyrighted data without authorisation is infringement, then many of the most commercially valuable AI systems in existence were built on an illegitimate foundation.
These two problems the authorship problem in the outputs and the infringement problem in the inputs are analytically distinct but practically interconnected. They will be examined in turn.
The Indian Statutory Framework Section 2(d)(vi) and the Computer-Generated Work Provision
The Indian Copyright Act, 1957 contains a provision that is, in the international context, unusually specific about the authorship of computer-generated works. Section 2(d)(vi) provides that in the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be the person who causes the work to be created.
This provision was introduced by the Copyright (Amendment) Act, 1994 a quarter century before the current generation of generative AI systems came into existence. Its context was the relatively modest form of computer-assisted creation that existed at the time: programs that generated reports, databases that produced outputs based on structured queries, algorithmic composition tools that created music from rule-based systems. The provision was designed to ensure that works generated through fully automated processes where no individual human author could be identified as the expressive source could nonetheless attract copyright protection and vest that protection in the person who had set the process in motion.
Section 2(d)(vi) offers a framework for allocating authorship in AI-generated works: the author is the person who causes the work to be created. Applied to a user who prompts a generative AI system to produce a novel, an image or a piece of music, the provision would suggest that the user who causes the system to create the output by providing the prompt and parameters is the author of the resulting work.
This interpretation has a certain practical appeal. It preserves the principle that copyright vests in a human being the person who caused the creation while accommodating the reality that the immediate creative act was performed by a machine. It provides a clear and administratively workable rule for copyright ownership. And it aligns with the general principle that the person who bears the creative and financial initiative for a work’s creation is its author.
However, Section 2(d)(vi) was not drafted with the current architecture of generative AI in mind and its application to modern AI outputs raises questions that its drafters could not have anticipated. The phrase “causes the work to be created” is ambiguous in the AI context. Does a user who types “write me a short story about a detective in Mumbai” into a large language model “cause” the resulting story to be created in a meaningful sense? The user has specified a general topic and format but has made no expressive choices about the specific words, sentences, narrative structure, character development or any other element that constitutes the literary expression of the work. The creative choices were made, to the extent that the word “choices” can be applied to a statistical system, by the AI model.
The Indian Copyright Office has not yet issued guidance on the application of Section 2(d)(vi) to large language model outputs, diffusion model images or other forms of modern generative AI output. No Indian court has directly ruled on the question. The provision remains the primary statutory hook for any analysis of AI authorship in India, but its adequacy for the task is genuinely questionable.
The Originality Requirement and AI-Generated Works
Even if Section 2(d)(vi) resolves the authorship question in favour of the human prompter, a second obstacle to copyright protection for AI-generated works remains: the originality requirement. Copyright subsists in original literary, dramatic, musical and artistic works under Section 13(1)(a). For a work to be original, it must originate from its author it must be the product of the author’s independent creative effort rather than a copy of something else.
The Supreme Court’s analysis of originality in Eastern Book Company v. D.B. Modak (2008) established that Indian copyright law requires not mere labour but a “minimal degree of creativity” some creative choice or judgment exercised by the author in creating the work. This standard, though deliberately modest, requires that the creativity be the author’s own that it reflect the author’s individual intellectual creation.
Applied to AI-generated works, the originality analysis becomes complex. The creative choices that determine the character of the output the word selection, the syntactic structures, the compositional decisions, the thematic development are made by the AI system based on its learned statistical model. The human user who provided the prompt has not made these choices. If originality requires that the creativity be the human author’s own and the creative choices in an AI-generated work are made by a non-human system, then the human user’s contribution however valuable as a creative instigation may not satisfy the originality requirement for the resulting work.
The question can be refined by considering the range of human input. Where a user provides an extremely detailed and specific creative brief a prompt that specifies not just the general topic but the specific narrative structure, the particular stylistic approach, the precise aesthetic choices and the specific expressive elements to be used the human’s creative contribution is substantial and the AI’s role is closer to that of an instrument for executing a detailed human creative vision. In this scenario, the argument that the human author satisfies the originality requirement is relatively strong. Where a user provides a brief and general prompt “write a poem about autumn” and receives a polished output that reflects the AI’s learned patterns rather than any specific human creative direction, the originality argument is much weaker.
This spectrum of human creative contribution suggests that the copyright question for AI-generated works is not binary but graduated. Works produced with substantial human creative direction may attract copyright; works produced with minimal human input may not. The difficulty is that the current statutory framework provides no mechanism for this kind of graduated analysis it requires a binary determination of authorship and originality that does not map well onto the reality of human-AI collaborative creation.
Comparative Approaches International Perspectives
The absence of settled Indian law on AI-generated works makes comparative analysis particularly valuable. The major copyright jurisdictions have adopted different approaches to this question, reflecting different underlying philosophies about the relationship between authorship, creativity and copyright protection.
The United States Copyright Office has addressed the question directly and has taken a position that is currently the most developed in any major jurisdiction. The Copyright Office has stated that it will not register copyright in works produced by a machine without creative input from a human author. In the case of AI-generated artworks, the Copyright Office has examined applications on a case-by-case basis, registering works where the human author’s creative contributions in selecting, arranging or modifying AI outputs are sufficient to constitute original human authorship, but refusing registration where the human’s role was limited to providing a prompt and accepting whatever the AI generated.
The landmark American decision in Thaler v. Perlmutter (2023) addressed the question directly when the developer of the DABUS AI system sought registration of an image generated entirely by the system without human creative input. The federal district court upheld the Copyright Office’s refusal to register the work, holding that copyright requires human authorship and that a work generated autonomously by an AI system without human creative contribution cannot be registered. The Court’s reasoning was categorical: copyright has never stretched to protect works produced by non-human actors and this principle applies to AI just as it applies to animals or natural forces.
The European Union’s approach to AI-generated works has been shaped by the civil law tradition’s emphasis on the author’s personality as the basis of copyright. The EU’s copyright framework, derived from the Berne Convention and implemented through harmonising directives, requires that a work reflect the author’s own intellectual creation a standard that, as the Court of Justice of the European Union has clarified in decisions on photographs and other borderline cases, requires that the author expresses their creative abilities by making free and creative choices. An AI system that does not make free creative choices in the human sense cannot satisfy this standard and EU law would not, on its current principles, recognise autonomous AI output as a copyrightable work.
The United Kingdom takes a distinctive approach. Section 9(3) of the Copyright, Designs and Patents Act, 1988 provides that in the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken. This provision which is substantially similar to India’s Section 2(d)(vi) means that UK law does recognise copyright in computer-generated works, with authorship vesting in the person who makes the necessary arrangements for the work’s creation. The term of protection for computer-generated works under UK law is fifty years from creation, shorter than the standard life-plus-seventy term.
China’s copyright authority issued a significant ruling in 2023 in which an AI-generated image was found to attract copyright protection on the basis that the human user who selected and refined the AI’s outputs through iterative prompting had made sufficient creative choices to be the author of the resulting image. This ruling aligned Chinese law with a generous approach to human-AI collaborative creation but was limited in its reasoning to cases where the human contribution was meaningful and iterative.
These comparative approaches reveal a spectrum from total exclusion of AI outputs from copyright protection at one end, through case-by-case assessment of human creative contribution, to the UK and Indian provisions that permit copyright in computer-generated works to vest in the person who causes or arranges their creation. India’s statutory position Section 2(d)(vi) is closest to the UK approach, but its application to modern generative AI remains undeveloped by the courts.
The Training Data Question Infringement at the Input Stage
The question of whether training generative AI systems on copyrighted works constitutes infringement of the copyright in those works is analytically prior to the output question and commercially even more consequential. The major AI companies OpenAI, Google, Meta, Stability AI and others have trained their models on datasets that include large quantities of copyrighted text, images, music and code. These datasets were assembled largely without the consent of the copyright holders and without payment of licensing fees.
Under Indian copyright law, the training of an AI model on copyrighted works requires the reproduction of those works they must be copied into a training dataset and processed by the model’s training algorithm. Reproduction of a copyrighted work without the owner’s authorisation is infringement under Section 51 of the Act. The question is whether any exception under Section 52 permits this reproduction.
The most relevant potential exception is the research and private study exception under Section 52(1)(a), which permits fair dealing with any work for purposes of private or personal use including research. If the training of an AI model constitutes research within the meaning of this provision and if the training is conducted for non-commercial purposes, a fair dealing defence might be argued. However, the commercial character of most AI training which is conducted by large technology companies for the purpose of creating commercially valuable AI systems makes the research exception unavailable. The fair dealing exception does not extend to commercial exploitation and AI model training conducted for profit cannot plausibly be characterised as fair dealing for private research.
The absence of any general “text and data mining” exception in the Indian Copyright Act is a significant gap. Several other jurisdictions the United Kingdom, Japan, the European Union through the DSM Directive have introduced specific exceptions permitting the reproduction of copyrighted works for text and data mining purposes, including for the purpose of training AI systems. The UK exception under Section 29A of the Copyright, Designs and Patents Act, 1988 permits text and data mining for any purpose, including commercial research. The EU DSM Directive Article 4 permits text and data mining for commercial purposes unless rights holders have expressly reserved their works from such use.
India has no equivalent provision. The Indian Copyright Act contains no text and data mining exception and none of the existing exceptions under Section 52 can be comfortably stretched to cover the commercial training of AI systems on copyrighted datasets. This creates a significant legal risk for AI companies operating in India that train their models on data that includes Indian copyrighted works and a corresponding rights enforcement opportunity for Indian copyright holders whose works have been used without consent.
The litigation landscape on this question is developing rapidly in the United States. The New York Times filed suit against OpenAI and Microsoft in December 2023 alleging that its articles were used to train ChatGPT without authorisation or compensation. Getty Images filed suits against Stability AI in both the United Kingdom and United States alleging that its image database was scraped for AI training without a licence. Universal Music Group and other major record labels sued AI music generation companies for training on copyrighted sound recordings. These cases are in various stages of litigation and their outcomes will inform the global legal landscape for AI training, including the interpretation of Indian law.
The question of whether the patterns learned during training the statistical weights encoded in a model’s parameters themselves constitute infringing reproductions of the training works is a further dimension of the input question. If the model has, in effect, memorised elements of its training data in a form that can be reproduced on demand, the case for infringement at the training stage is stronger. If the model has genuinely abstracted statistical patterns without retaining reproductions of specific works, the infringement argument is weaker. The technical reality is complex and context-dependent: models do sometimes reproduce training data verbatim when prompted in particular ways and the extent to which this constitutes reproduction of a “substantial part” of a copyrighted work within Section 51 is a question that Indian courts will eventually need to address.
The Output Infringement Question When AI Outputs Reproduce Existing Works
Independently of the training data question, the outputs of AI systems may themselves infringe copyright if they reproduce a substantial part of a copyrighted work. This can occur in several ways.
The most straightforward case is direct reproduction where an AI system, when prompted in a particular way, generates output that reproduces verbatim or near-verbatim passages from works in its training data. This phenomenon has been demonstrated empirically for large language models, which can be induced to reproduce recognisable passages from training texts under certain prompting conditions. Where the reproduced passage constitutes a substantial part of the original work assessed qualitatively rather than merely quantitatively, as Indian and international courts consistently do the reproduction constitutes infringement regardless of whether it was generated by a human or a machine.
A more complex case arises where the AI output does not reproduce specific text or images but generates content that is substantially similar to a copyrighted work in its overall structure, narrative pattern, compositional approach or stylistic character. The idea-expression dichotomy the principle that copyright protects expression but not ideas places limits on the scope of protection, but courts have recognised that highly specific structural and expressive choices can attract protection even where no verbatim reproduction has occurred. The standard Indian test for infringement, derived from the Supreme Court’s decision in R.G. Anand v. Deluxe Films (1978), asks whether the work as a whole creates the same impression in the mind of a reasonable person as the original. Applied to AI outputs, this test requires case-by-case assessment that courts are not yet systematically equipped to perform.
The stylistic similarity question is particularly acute for AI image generation. Diffusion models such as Stable Diffusion, Midjourney and DALL-E generate images that reproduce the aesthetic style of specific artists with remarkable fidelity when prompted with the artist’s name. The question of whether the generation of an image “in the style of” a named artist infringes the copyright in that artist’s works is contested. The prevailing view is that style is an idea rather than an expression that copyright does not protect the distinctive visual approach of an artist, only the specific works in which that approach is embodied. On this view, the generation of an image in the style of a named artist is not copyright infringement, even if the artist finds the practice objectionable and commercially threatening.
However, where the AI-generated image so closely resembles a specific copyrighted work that a reasonable observer would consider them substantially similar where the “style” in question amounts to a reproduction of specific expressive choices rather than a general aesthetic approach the infringement analysis may produce a different result. This boundary between protected style and unprotected expression is not clearly drawn in Indian law and its application to AI-generated content will require judicial development.
Moral Rights and AI-Generated Works
The moral rights framework under Section 57 of the Copyright Act the right of paternity and the right of integrity presents distinctive questions in the AI context.
The right of paternity is the right to be identified as the author of a work. Where an AI system is used to generate a work and the human user publishes that work under their own name without disclosure of AI involvement, the paternity right question is straightforward the human is identified as the author, accurately or not depending on the degree of human creative contribution. Where the AI-generated work is published without any human attribution as, for example, where an AI system autonomously generates and publishes content the paternity right has no obvious beneficiary, because the system cannot hold rights and no human author is identified.
The right of integrity the right to object to distortion, mutilation or modification of a work prejudicial to the author’s honour or reputation is even more difficult to apply in the AI context. Integrity rights under Section 57 are available to the “author” of a work. If an AI system is not an author, it cannot hold an integrity right. If the human user who prompted the AI is treated as the author under Section 2(d)(vi), they theoretically hold the integrity right but the integrity right protects the author’s personal honour and reputation, concepts that have a different content when applied to a person who directed rather than created the expressive content of the work.
The moral rights question has direct practical implications for artists and authors whose works are used to train AI systems. An artist whose distinctive style is reproduced by an AI system even without reproducing specific works verbatim may argue that the AI’s generation of works attributed to them or works that misrepresent their expressive approach, violates their integrity right. This argument has not been tested in Indian courts and its success would require a significant extension of Section 57 beyond its current judicial interpretation.
The Ownership Question in Commercial Practice Who Gets the Benefit
In the commercial practice of AI-generated content, the ownership question is typically resolved by the terms of service of the AI platform rather than by copyright statute. The major AI platforms take different positions on the ownership of user-generated outputs.
OpenAI’s terms of service for ChatGPT and DALL-E assign to the user all rights, title and interest in the content they generate using the tools, to the extent that OpenAI holds any such rights. This assignment is contractually effective between the parties but cannot create copyright where none subsists by law if AI-generated outputs attract no copyright protection under Indian law, there are no rights to assign. Midjourney’s terms of service take a different approach, retaining rights for the company in certain commercial contexts while granting users licences for personal use. Stability AI’s terms permit broad use of outputs with limited restrictions.
These contractual arrangements operate in a legal vacuum in India in the absence of statutory clarity on whether AI-generated outputs attract copyright at all and in whom any such copyright vests, the contractual terms of AI platforms are the primary governance mechanism for the ownership of AI outputs in commercial transactions. This is an unstable and unsatisfactory position that creates legal uncertainty for every business that uses AI-generated content in its commercial activities.
For Indian companies that commission AI-generated content for commercial use marketing materials, product imagery, training datasets, legal documents the absence of clear copyright ownership creates practical risks. If the AI-generated output does not attract copyright protection, it may be freely copied by competitors without legal redress. If it does attract copyright and that copyright vests in the AI company by operation of the platform’s terms of service rather than in the commissioning user, the user’s ability to enforce rights against infringers is compromised.
AI as a Tool Versus AI as the Author The Critical Distinction
A conceptually important distinction that has emerged in international discussions of AI and copyright is the distinction between AI used as a creative tool and AI operating as the autonomous creative agent. This distinction tracks the spectrum of human creative contribution discussed above and is the key to any graduated approach to AI copyright.
Where AI is used as a tool in the service of human creative vision as a powerful instrument for the realisation of the human author’s specific expressive choices the copyright analysis is relatively straightforward. The photographer who uses AI-powered editing software to process their images, the composer who uses an AI-assisted notation tool to transcribe their musical ideas, the novelist who uses an AI grammar assistant to refine their prose all of these are human authors using AI tools and the resulting works reflect their authorship in the traditional sense. Copyright vests in the human author without difficulty.
Where AI operates with substantial autonomy where the human’s contribution is limited to a general prompt and the system makes all the specific expressive decisions that constitute the creative character of the work the human’s role is closer to that of a commissioner than an author and the copyright analysis becomes genuinely difficult. The work does not reflect the human’s personal intellectual creation in the sense that copyright law has traditionally required.
The boundary between these poles is not fixed it shifts with the specificity of the human’s input, the autonomy of the AI system and the nature of the creative domain. It is also technologically dynamic as AI systems become more capable and prompting becomes more sophisticated, the effective creative contribution of the human prompter may increase or decrease depending on how the interaction is structured.
Indian courts, when they eventually address these questions, will need to develop a framework for this spectrum. The statutory provision in Section 2(d)(vi) the person who causes the work to be created provides a starting point but not a complete answer. The content of “causing” a work to be created and the minimum level of human creative direction required to make the causal relationship sufficient for authorship, will need to be elaborated through judicial development.
The Performers’ Rights Dimension AI-Synthesised Vocal Performances
The use of AI to synthesise vocal performances in the style of specific human performers a technology now widely available and commercially deployed raises questions under the performers’ rights provisions of the Copyright Act that deserve separate attention.
Section 38 of the Act provides performers with exclusive rights including the right to make a sound recording of their performance and the right to communicate their performance to the public. Section 38B provides performers with moral rights including the right to be identified as the performer and the right to object to modifications of their performance prejudicial to their reputation.
Where an AI system is used to generate a vocal performance that sounds like a specific human performer reproducing their vocal timbre, style and expressive characteristics without using any recording of their actual voice the question is whether this engages the performer’s rights under Section 38. The performers’ rights provisions protect “performances” a term defined in Section 2(q) as any visual or acoustic presentation made by a performer. An AI-synthesised vocal performance is not a performance by the human performer in any conventional sense it is a simulation of how they might perform, generated by a system trained on recordings of their actual performances.
The performers’ rights provisions do not directly address this scenario and their extension to AI simulation of a performer’s voice would require either a creative reading of the existing provisions or, more probably, legislative amendment. The moral rights of performers under Section 38B particularly the integrity right might offer one avenue: if an AI system generates a performance attributed to a real performer or so closely resembling their voice as to create that impression, the performer might argue that the attribution or misattribution damages their reputation. But this argument has not been tested in Indian courts and its success is uncertain.
The Case for Legislative Reform
The inadequacy of the existing Copyright Act framework for the governance of AI-generated works is apparent from the analysis above. Section 2(d)(vi), while a useful starting point, was not designed for the current generation of generative AI and cannot answer the specific questions that modern AI-generated content raises. The absence of a text and data mining exception leaves AI training on an uncertain legal footing. The absence of guidance on the minimum human creative contribution required for copyright protection of AI outputs creates commercial uncertainty. The performers’ rights framework does not address AI voice synthesis. And the moral rights provisions are difficult to apply to works without a human author in the traditional sense.
The legislative response required is not simple, because the policy questions involved are genuinely complex. Extending full copyright protection to AI-generated outputs on the same terms as human-authored works would significantly expand the scope of copyright and potentially disadvantage human creators in the market for creative content. Denying all copyright protection to AI-generated outputs would leave a large and growing category of commercially valuable content unprotected, creating free-rider problems and reducing the incentive to invest in AI-assisted creative production. A middle path recognising reduced or qualified protection for AI-generated works or requiring minimum levels of human creative contribution for full protection is intellectually attractive but practically difficult to administer.
The text and data mining question is more tractable. India should consider following the European and British approach and introducing a specific exception for non-consumptive use of copyrighted works for machine learning and AI training purposes, subject to rights holder opt-out mechanisms that allow those who object to have their works excluded from training datasets. This would provide a legal basis for AI training while preserving rights holders’ ability to control the commercial exploitation of their works.
A registration or disclosure requirement for AI-generated commercial content requiring commercial publishers of AI-generated works to disclose the AI involvement would serve both the transparency interests of the public and the attribution interests of human creators who might otherwise find their audiences confused about the origin of content they encounter.
Implications for the Indian Creative and Technology Industries
The AI copyright question has direct implications for two of India’s most important industries the creative industries that produce film, music, literature and art and the technology industry that develops, deploys and exports AI systems.
For the Indian creative industries, the unresolved status of AI-generated works creates both threats and opportunities. The threat is that AI systems trained on Indian creative works Bollywood music, regional literature, Indian visual art traditions may generate commercially competitive content without compensating the creators of the works on which they were trained. The opportunity is that India’s position as a major producer of creative content gives its rights holders significant leverage in global discussions about AI training data licensing and that a clear Indian legal framework protecting those rights would strengthen that leverage.
For the Indian technology industry, which includes several major AI companies and a large and growing AI services sector, the absence of a text and data mining exception and the unclear copyright status of AI outputs creates legal risk that affects competitiveness. Indian AI companies that train their models on data that includes Indian copyrighted works operate in a state of legal uncertainty that their counterparts in jurisdictions with clear text and data mining exceptions do not face. Regulatory clarity would benefit the technology industry while protecting the legitimate interests of rights holders.
Conclusion
The question of who owns AI-generated works is, at present, unanswered under Indian law. Section 2(d)(vi) of the Copyright Act provides a framework the person who causes the work to be created is the author of a computer-generated work but its application to modern generative AI systems is uncertain and contested. The originality requirement creates a further obstacle: AI-generated works produced with minimal human creative direction may not satisfy the requirement that the work reflect the author’s own intellectual creation. The training data question is entirely unaddressed by the existing statutory exceptions. The moral rights and performers’ rights dimensions are similarly underdeveloped.
The resolution of these questions cannot be left entirely to the courts. Judicial interpretation of the existing provisions can develop the law incrementally, but the structural gaps the absence of a text and data mining exception, the inadequacy of the computer-generated work provision for modern generative AI, the silence on AI voice synthesis and AI-generated performances require legislative intervention. India’s position as both a major producer of creative content and a major developer and exporter of AI technology makes the stakes of this legislative choice particularly high.
The international experience particularly the United States Copyright Office’s approach, the European Union’s DSM Directive and the developing litigation landscape provides valuable reference points. But India’s framework must ultimately reflect Indian policy priorities: the protection of the Indian creative community, the promotion of the Indian AI industry and the maintenance of a copyright system that serves the public interest in access to creative and informational works.
What is clear is that the question cannot be indefinitely deferred. Generative AI is not an emerging technology it is a current one, deployed at scale in every sector of the economy, generating content that is being published, sold and licensed today. The legal framework that governs the ownership of that content is needed now and its absence imposes real costs on both creators and the industry that serves them.
References
- The Copyright Act, 1957, Sections 2(d)(vi), 2(o), 13, 14, 17, 38, 38A, 38B, 51, 52, 57, 65A, 65B https://copyright.gov.in/Documents/CopyrightRules1958.pdf
- The Copyright (Amendment) Act, 2012 https://copyright.gov.in/Documents/Amendment_Act2012.pdf
- Eastern Book Company v. D.B. Modak, (2008) 1 SCC 1 (Supreme Court of India) https://indiankanoon.org/doc/1023365/
- R.G. Anand v. Deluxe Films & Ors., AIR 1978 SC 1613 (Supreme Court of India) https://indiankanoon.org/doc/595730/
- Thaler v. Perlmutter, Case No. 22-1564 (D.D.C. 2023) https://www.courtlistener.com/docket/63356632/thaler-v-perlmutter/
- U.S. Copyright Office Copyright and Artificial Intelligence Policy (2023–2024) https://www.copyright.gov/ai/
- EU Directive on Copyright in the Digital Single Market (DSM Directive), 2019/790, Articles 3–4 https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32019L0790
- UK Copyright, Designs and Patents Act, 1988, Section 9(3), Section 29A https://www.legislation.gov.uk/ukpga/1988/48/contents
- Berne Convention for the Protection of Literary and Artistic Works, Article 2 https://www.wipo.int/treaties/en/ip/berne/
- WIPO Conversation on Intellectual Property and Artificial Intelligence https://www.wipo.int/about-ip/en/artificial_intelligence/
- WIPO Performances and Phonograms Treaty, 1996 https://www.wipo.int/treaties/en/ip/wppt/
- TRIPS Agreement https://www.wto.org/english/docs_e/legal_e/27-trips.pdf
- Indian Performing Right Society v. Eastern India Motion Pictures Association, AIR 1977 SC 1443 https://indiankanoon.org/doc/553674/
- Amarnath Sehgal v. Union of India, 117 (2005) DLT 717 (Delhi High Court) https://indiankanoon.org/doc/1402532/
- Copyright Office of India https://copyright.gov.in
Compulsory Licensing Copyright Act Copyright Act 1957 Copyright Infringement Copyright Law Copyright Rules Deceptive Similarity Descriptive Marks India Goodwill Indian IP Framework Indian Patent Law Indian Trademark Law Intellectual Property Law IP Law India Patent Claims Patent Enforcement Patent Infringement Patent law Patent Revocation Patent Rule Patents Act 1970 Pharmaceutical Patents Section 3 Section 29 The Patent Act 1970 Trademark Infringement Trademark Registration Trade Marks Act 1999 Trade Marks Rules 2017 TRIPS Compliance