Machines That Create, Laws That Don’t: Intellectual Property Protection for AI-Generated Art and Text Under the Indian Copyright Act, 1957
Published in: Law Vaani | Category: Intellectual Property Law
Part I: Abstract and Introduction
The emergence of generative artificial intelligence (AI) has inaugurated a new era of creative production in which machines independently author visual art, draft literary texts, compose music, and generate vast quantities of content at speeds and scales previously unimaginable. Platforms such as Midjourney, DALL-E, Stable Diffusion, and ChatGPT have made AI-assisted creativity accessible to millions of users across India and the world, fundamentally disrupting the traditional relationship between human effort and creative output. Works of apparent aesthetic sophistication are now produced within seconds, at negligible cost, and without any conventional act of human artistic labour. Yet as these tools proliferate, they expose a profound and structurally consequential gap within India’s intellectual property framework: the Indian Copyright Act, 1957, enacted in a pre-digital era, does not explicitly contemplate the existence of an artificial intelligence as a creative entity. The question that confronts courts, legislators, and jurists is deceptively simple yet legally formidable — when a machine creates, who owns what it makes, and can the law protect it at all?
This article undertakes a systematic legal analysis of the protectability of AI-generated art and text under the Copyright Act, 1957. It examines the statutory foundations of copyright in India — specifically Sections 2(d), 13, and 17 — and interrogates whether existing provisions adequately assign authorship, ownership, and protection to works created autonomously or semi-autonomously by AI systems. It further analyses the doctrine of fair dealing under Section 52, the landmark litigation in ANI Media Pvt. Ltd. v. OpenAI Inc. & Anr. (CS(COMM) 1028/2024) before the Delhi High Court, and the path-breaking policy interventions of the Department for Promotion of Industry and Internal Trade (DPIIT) through its 2025 expert committee and December 2025 Working Paper on Generative AI and Copyright. A comparative survey of international frameworks — including the United Kingdom, the United States, China, and Japan — is also undertaken to identify best practices. The article concludes with independent reform recommendations aimed at constructing a legally coherent and technologically responsive framework for the protection of AI-generated creative works in India.
Part II: Legal Framework — The Copyright Act, 1957 and the Anatomy of Protection
A. Foundational Structure of Copyright in India
The Copyright Act, 1957, as amended most recently in 2012, constitutes the primary legislative instrument governing the protection of original creative works in India. Section 13(1) defines the classes of works in which copyright subsists, namely original literary, dramatic, musical, and artistic works, as well as cinematographic films and sound recordings. The word ‘original’ is critical, as it sets the threshold for protection. While the Act does not explicitly define originality, Indian courts have interpreted it to require at minimum a modicum of creativity reflecting the author’s own intellectual effort, as elaborated by the Supreme Court in Eastern Book Company v. D.B. Modak, (2008) 1 SCC 1, which rejected the mere ‘sweat of the brow’ standard in favour of requiring a genuine creative contribution from the author.
Section 14 defines copyright as a bundle of exclusive economic rights, including the right to reproduce, publish, adapt, and translate a work. Section 17 vests first ownership of copyright in the author, except where the work is created in the course of employment or upon commission, in which case the employer or commissioner holds ownership. The interplay of these foundational provisions creates the analytical edifice upon which questions of AI-generated authorship must now be carefully examined.
B. The Pivotal Definition: Section 2(d) and the Question of Authorship
The most consequential provision in the context of AI-generated works is Section 2(d) of the Copyright Act, 1957, which defines ‘author’ across six distinct categories. For literary, dramatic, and musical works, the author is the person who creates the work. For artistic works, it is the artist. The provision most directly relevant to the AI debate is Section 2(d)(vi), introduced by the Copyright (Amendment) Act, 1994, which defines the author of a ‘computer-generated work’ as ‘the person who causes the work to be created.’
This sub-clause was designed for an era of deterministic computing, where a human programmer directed a computer through defined instructions to produce a predictable output. The legislative intent was clear: a human being stood behind every computer-generated creative act. However, this formulation is structurally ill-suited for modern generative AI, where models trained on billions of data points produce outputs with minimal or entirely absent human authorial direction. The interpretive dilemma under Section 2(d)(vi) is threefold: first, whether an AI system qualifies as a ‘computer’ under Section 2(ffb), which broadly defines a computer as ‘any electronic or similar device having information processing capabilities’; second, whether a brief user-supplied prompt constitutes ‘causing’ a work to be created within the meaning of the provision; and third, where multiple actors — the AI developer, the training data provider, and the end-user — each exercise partial causal influence, who is the true causal author. Indian copyright jurisprudence has not yet resolved any of these questions through direct judicial pronouncement.
C. The Originality Problem: Creativity Without a Creator
Even if the authorship question were resolved in favour of a human claimant, AI-generated works face an independent and formidable challenge on the ground of originality. Section 13 of the Act requires that a work be ‘original’ to attract copyright protection. In Eastern Book Company v. D.B. Modak, the Supreme Court held that a work must reflect the author’s intellectual creativity, displaying the stamp of the author’s personality. For a work generated autonomously by an AI in response to a generic prompt such as ‘draw a sunset over the Ganges’ or ‘write a poem about freedom,’ it is virtually impossible to demonstrate that the resulting output bears the stamp of any individual human creative personality.
The challenge is compounded by the statistical nature of generative AI output. Large language models and image generators produce outputs by calculating probabilistic distributions over vast training datasets, optimising for patterns that resemble human-created work. There is no intentionality, aesthetic vision, or emotional investment in the AI’s creative process. The doctrine of minimal creativity, drawn from the US Supreme Court’s decision in Feist Publications Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), requires at minimum a spark of independent creative judgment. An autonomous AI system, by definition, cannot supply that spark. Accordingly, Indian law as presently constituted does not recognise purely AI-generated content as copyrightable unless a human can be identified as its substantive author or controller.
Part III: Main Analysis — Judicial Developments and the Precedent Landscape
A. ANI Media Pvt. Ltd. v. OpenAI Inc. & Anr. (CS(COMM) 1028/2024)
The most significant judicial development in Indian AI-copyright law to date is the ongoing case of ANI Media Pvt. Ltd. v. OpenAI Inc. & Anr. (CS(COMM) 1028/2024), being heard before Justice Amit Bansal of the Delhi High Court. The suit was filed in November 2024 by Asian News International (ANI), alleging that OpenAI had unlawfully ingested ANI’s proprietary news articles — including paywalled content accessible only to subscribers — to train its large language model, ChatGPT. ANI further alleged that ChatGPT reproduced verbatim excerpts from its articles in responses to user queries and falsely attributed fabricated news stories to ANI, causing significant reputational harm and spreading misinformation. ANI sought a restraining injunction and claimed damages of Rs. 2 crore. On October 3, 2024, ANI had offered OpenAI a formal licensing agreement, which OpenAI refused to accept — a refusal that ANI submitted demonstrated deliberate disregard for intellectual property obligations.
OpenAI, in its 31-page reply filed in February 2025, contended that it relied exclusively on publicly available data and that such use fell within fair dealing under Section 52 of the Copyright Act. It also challenged the Delhi High Court’s territorial jurisdiction, arguing its servers were located outside India. The court appointed Professor Arul George Scaria of the National Law School of India University as amicus curiae. After detailed hearings through 2025 and into 2026, the court reserved judgment — a ruling expected to function as the inaugural authoritative interpretation of AI and copyright in India, carrying both domestic and global precedential significance.
The case raises three interrelated questions of cardinal legal importance: whether ingesting and storing copyrighted material for AI training constitutes ‘reproduction’ under Section 14; whether AI-generated responses derived from such training data independently infringe copyright; and whether such use is justified under Section 52’s fair dealing provisions. The court’s answers will shape the commercial and legal architecture of AI development in India for years to come.
B. The Fair Dealing Doctrine Under Section 52: A Contested Boundary
Section 52 of the Copyright Act, 1957 enumerates acts that do not constitute infringement, including fair dealing for research, private study, criticism, review, and reporting of current events. Unlike the American fair use doctrine, which employs a flexible multi-factor balancing test, India’s fair dealing exception is exhaustive and categorical: a use must fall squarely within a listed category, and courts retain no residual discretion to recognise new categories of permissible use. The DPIIT’s expert committee, constituted in April 2025, observed that the research exception under Section 52 is limited to non-commercial purposes. Since AI companies train models on copyrighted data with clear commercial intent, the DPIIT has unequivocally rejected any blanket fair dealing defence for such entities. This position aligns India’s emerging policy stance with the EU’s 2019 Digital Single Market Directive, which explicitly distinguishes between non-commercial research use and commercial-scale data extraction — a distinction that Indian law currently fails to make, and which must urgently be addressed through legislative intervention.
It bears noting that the fair dealing provisions under Section 52, as presently drafted, were conceived to address the incidental and bounded use of copyrighted material by individual human researchers, critics, and journalists. They were never intended to accommodate the systematic and industrial-scale ingestion of copyrighted datasets by commercial technology enterprises for the purpose of training machine learning systems that will themselves generate and distribute content commercially. Applying Section 52 to AI training without amendment would be an act of interpretive overreach that fundamentally distorts the provision’s legislative intent and threatens the economic foundations of India’s creative industries.
C. Comparative Jurisprudence: Lessons from International Frameworks
A comparative survey reveals the diversity of international approaches to AI-generated copyright. In the United Kingdom, Section 9(3) of the Copyright, Designs and Patents Act, 1988 vests authorship in ‘the person by whom the arrangements necessary for the creation of the work are undertaken’ — functionally analogous to Section 2(d)(vi) of the Indian Act. The US Court of Appeals for the District of Columbia Circuit, in Thaler v. Perlmutter (March 2025), affirmed that works created solely by AI without human authorship cannot attract copyright protection, confirming the longstanding position of the US Copyright Office.
China’s Beijing Internet Court, in a landmark 2023 ruling, held that an AI-generated image could attract copyright where the human user demonstrated sufficient creative input through iterative prompting and aesthetic selection — a ‘meaningful human contribution’ test that offers a useful doctrinal reference for Indian courts seeking to calibrate the threshold of protectable human involvement. Japan has adopted a broad TDM exception permitting commercial use of copyrighted works for AI training. The DPIIT has studied these varied international models but declined to wholesale adopt any single approach, instead proposing a balanced hybrid framework tailored to India’s distinctive legal traditions and economic development priorities.
Part IV: Critical Analysis — Regulatory Lacunae and the Path to Reform
A. The Structural Inadequacy of Section 2(d)(vi) for Generative AI
The central legislative deficiency is that Section 2(d)(vi), as drafted in 1994, was designed for a paradigm of human-directed computing that bears little resemblance to contemporary generative AI. In 1994, computer-generated works referred to outputs produced by programs humans coded and controlled at every step. Generative AI inverts this relationship entirely: the model learns autonomously from training data and produces outputs that no human could have precisely predicted or predetermined. The multi-party structure of AI creation further complicates attribution across the entire value chain. Developers who construct and train the model, businesses that deploy it as a commercial service, and end-users who supply prompts each exercise some degree of causal influence over the final output. If the prompt-giver is deemed the author, a two-word prompt would generate copyright in sophisticated artistic work, diluting the originality requirement to the point of absurdity. If the developer is the author, a single entity could claim copyright over millions of daily user-generated outputs, producing a monopolistic outcome wholly inconsistent with copyright law’s foundational policy objectives of promoting creativity and protecting individual expression.
B. The Training Data Problem: Infringement at the Foundation
An equally critical dimension concerns the legality of using copyrighted works to train AI models. Every sophisticated generative AI system is trained on massive datasets scraped from the internet — published literary works, journalism, academic papers, visual art, and music — much of which is protected by copyright. The act of reproducing and processing these works in training pipelines potentially constitutes reproduction within the meaning of Section 14. The DPIIT expert committee’s Working Paper on Generative AI and Copyright, released in December 2025, assessed several regulatory models including blanket licensing, TDM exceptions with opt-out rights, and extended collective licensing. The committee proposed introducing a statutory licensing regime under which AI developers pay fair compensation to copyright holders through a collectively administered royalties fund. This ‘compensation-for-access’ framework seeks to balance creators’ legitimate economic rights against India’s imperative to remain globally competitive in AI development and to attract investment in the AI sector.
C. The Spectrum of Human Involvement in AI-Assisted Works
Not all AI-generated content exists at the extreme of full autonomy. A spectrum of human involvement characterises contemporary AI use: from fully autonomous generation responding to a minimal prompt, to AI-assisted creation where a human artist uses generative tools as creative instruments while exercising iterative judgment over hundreds of outputs before selecting and refining the final work. Indian courts will need to identify the threshold at which an AI-assisted work transforms into a copyrightable human creation. The DPIIT committee’s preliminary deliberations have proposed distinguishing between ‘AI-assisted’ works — where the human exercises meaningful creative control throughout the process — and ‘AI-generated’ works — where the machine operates with substantial autonomy. This distinction, if codified in statute, would give courts a workable and principled framework to resolve authorship disputes without necessitating a wholesale reconceptualisation of Indian copyright doctrine.
Part V: Conclusion — Reforms and the Road Ahead
The Indian Copyright Act, 1957 stands at the threshold of its most profound structural challenge since the digital amendments of 1994 and 2012. The current framework is demonstrably insufficient to address the complex authorship, ownership, and infringement questions raised by AI-generated art and text. Three critical reform imperatives emerge from this analysis, each of which demands urgent legislative and judicial attention.
First, Section 2(d) must be amended to introduce an explicit definition of ‘AI-generated work’ and a corresponding authorship framework. The preferred approach should legislatively distinguish between AI-assisted and AI-generated works. For AI-assisted works, authorship should vest in the human who exercises demonstrable and significant creative control over the final output, assessed by the nature and degree of that human’s creative investment. For fully autonomous AI-generated works, a sui generis protection regime should be introduced, with ownership potentially vesting in the entity that deployed the AI system for a limited and clearly defined term, thereby preventing the vacuum of an entirely unowned work while avoiding the overextension of full copyright monopoly to machine output.
Second, a statutory text and data mining exception should be introduced under Section 52, alongside a mandatory compensation mechanism channelled through a collectively administered copyright royalties fund. India should resist wholesale adoption of either the permissive Japanese model or the restrictive EU approach, instead crafting a distinctly Indian framework that permits AI training on publicly available copyrighted works subject to fair and transparent remuneration to all affected creators and rights holders.
Third, the forthcoming judgment in ANI Media v. OpenAI must be harnessed as the foundation for clear, principles-based judicial guidance on AI and copyright. The Delhi High Court has a historic opportunity to articulate India’s first authoritative interpretation of the Copyright Act’s application to generative AI. Whether its ratio decidendi turns on the scope of reproduction under Section 14, the limits of fair dealing under Section 52, or the territorial dimensions of AI infringement, the court’s reasoning will echo through every subsequent AI-copyright dispute in India. Intellectual property law must remain a living instrument that evolves in step with the societies and technologies it governs. Meeting this challenge requires both legislative courage and judicial wisdom in equal measure.
References
1. The Copyright Act, 1957 (Act 14 of 1957) — Sections 2(d), 2(ffb), 13, 14, 17, 52. Ministry of Law and Justice, Government of India.
2. The Copyright (Amendment) Act, 1994 — Introduction of Section 2(d)(vi) for computer-generated works.
3. ANI Media Pvt. Ltd. v. OpenAI Inc. & Anr., CS(COMM) 1028/2024, Delhi High Court (Judgment Reserved, 2026).
4. Eastern Book Company v. D.B. Modak, (2008) 1 SCC 1, Supreme Court of India.
5. Feist Publications Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), US Supreme Court.
6. Thaler v. Perlmutter, D.C. Circuit Court of Appeals (March 2025).
7. DPIIT, Working Paper on Generative AI and Copyright (Part I), December 8, 2025.
