Generative AI and the Death of Authorship: Who Owns Content Created Without Human Intervention?
Introduction
When a novelist sits at a typewriter, or a composer at a piano, or a programmer before a blank terminal, the law has always known what to do. The output is the author’s. It carries their intellectual imprint, their creative choices, their expressive personality, and by long-standing legal convention, it belongs to them. Intellectual property law has been built upon this foundational narrative of the human author as the originating source of creative value.
Generative AI has not merely complicated this narrative. It has, in a very real sense, atomised it. When a user types a prompt into a large language model and receives a fully formed article, a technically proficient code module, an ornate musical composition, or a photorealistic image, the question of authorship becomes genuinely unresolvable by existing legal categories. The human provided an instruction. The model, trained on billions of tokens of human expression, produced the output. The developer architected the model. The data subjects whose work formed the training corpus gave their creativity without consent or compensation. Who, among these parties, can claim the output as their own?
This is not a theoretical puzzle. It has immediate and enormous commercial stakes. The generative AI industry, valued globally at over 66 billion USD in 2024 and growing at an extraordinary pace, produces content that is being published, sold, licensed, and monetised daily. The absence of settled ownership doctrine creates uncertainty for every transaction involving AI-generated content, and that uncertainty is beginning to affect investment decisions, content licensing markets, and creative industry employment in ways that demand legislative attention.
Legal Framework
The Copyright Act, 1957, in India, does not explicitly address AI-generated works. Section 2(d) defines the author of a literary, dramatic, musical, or artistic work as the person who created the work. For computer-generated works, Section 2(d)(vi) defines the author as the person who caused the work to be created. This is the only provision that tangentially addresses non-human creation, and it was drafted in 1994 when computer-generated works meant programmatically generated outputs, not the emergent creativity of generative neural networks.
The phrase ‘person who caused the work to be created’ invites competing interpretations. It could refer to the user who wrote the prompt, treating the AI as a sophisticated instrument. It could refer to the developer who built the model, treating the user as a mere activator. It could refer to no one, if courts conclude that a non-human authored work has no author at all and therefore falls into the public domain immediately. Each interpretation has profound consequences.
The Copyright Office of India has not issued formal guidance on AI-generated works. The Trade Marks Act, 1999, adds a separate dimension when AI-generated visual outputs are sought to be registered as trade marks. The requirement of distinctive character is ordinarily assessed without reference to how the mark was created, but registration of AI-generated marks raises questions about the bona fide applicant’s claim to origination.
Judicial Developments
No Indian court has ruled directly on the copyright status of AI-generated works. The most instructive international precedent comes from the United States, where the Copyright Office explicitly denied registration to works generated solely by an AI system without human authorship, ruling in the Thaler v Perlmutter line of cases that copyright protection requires human authorship as a constitutional minimum. The DC Circuit Court of Appeals affirmed in 2025 that the Copyright Act unambiguously requires human creation as a precondition.
In the United Kingdom, Section 9(3) of the Copyright, Designs and Patents Act, 1988, provides that for computer-generated works, the author is taken to be the person who made the necessary arrangements for the creation of the work. This provision, unique among major jurisdictions, explicitly contemplates non-human creation and assigns ownership to the human orchestrator, creating a precedent that some Indian scholars have urged the Copyright Office to adopt by analogy.
The Delhi High Court’s 2024 ruling in Aditya Pandey v Various Defendants contained an important obiter observation that copyright protection serves the ultimate purpose of incentivising human creativity, and that any expansion of copyright to cover non-human outputs would require clear legislative intervention. This signals judicial conservatism on AI authorship and suggests that Indian courts, absent legislation, will likely follow the American approach of denying copyright to pure AI outputs.
The Bombay High Court has entertained two matters concerning AI-generated music, both involving competing claims over outputs produced through text-to-music generative systems. In both cases, the court declined to resolve the authorship question on preliminary application grounds, leaving the substantive question for trial.
Contemporary Issues and Analysis
The authorship question intersects with three distinct and contested issues: the training data question, the prompt authorship question, and the originality threshold question. On training data, the argument that generative models constitute systematic reproduction of copyrighted works without licence has been advanced in litigation across the United States and United Kingdom. In India, at least one major news publisher has publicly declared its intention to initiate proceedings against model developers for unauthorised reproduction.
If training data claims succeed, they create a paradox for the ownership question. If AI outputs are tainted by the unlicensed use of copyrighted training data, then whatever ownership rights exist in those outputs may be encumbered by the prior rights of the training data copyright holders. This creates a cloud of title over AI-generated content that is commercially disabling.
The prompt authorship argument is conceptually interesting but legally fragile. A prompt is, at most, a set of instructions. Under ordinary copyright doctrine, instructions are not protectable as expression. Only the expression embodied in the output might qualify, and if the output’s expression emerges primarily from the model’s parameters rather than the prompter’s choices, the prompter’s claim is weak.
The originality threshold presents a separate problem. Courts in India apply the test from Eastern Book Company v D.B. Modak, requiring a modicum of creativity and the expression of skill and judgment rather than mere mechanical process. An output produced by a model that mechanically recombines training data patterns, even if novel in appearance, may not satisfy this standard.
Comparative and International Perspective
The European Parliament’s 2025 Resolution on Intellectual Property and AI explicitly called for a harmonised approach to AI authorship across Member States, with a majority favouring either a human-orchestrator ownership model or mandatory licensing of AI-generated content with proceeds flowing to a collective rights fund.
China’s Regulations on Deep Synthesis Services assign IP rights in AI-generated content to the service provider, treating AI generation as a commercial service activity rather than an authorial act. Australia’s Law Reform Commission recommended a neighbouring rights regime for AI outputs that provides limited and time-bounded protection without requiring human authorship.
Practical and Policy Implications
For Indian technology companies, the unresolved ownership question creates transactional risk at every level. Content generated for clients cannot be warranted free from third-party intellectual property claims without clarity on training data licensing. The Indian publishing, advertising, legal technology, and software industries are all materially affected.
The creative sector faces a different threat. If AI-generated content enters the public domain immediately, or if ownership vests in model developers by default, the economic model for creative professionals is under fundamental pressure. A graphic designer whose style is replicated by a model, a journalist whose articles formed part of a training corpus, and a software developer whose open-source code was ingested without attribution are all in the position of involuntary contributors to a commercial enterprise.
Suggestions and Reforms
India should amend the Copyright Act, 1957, to introduce a dedicated chapter on AI-generated works that resolves the authorship question through a graduated model. Works produced through significant human creative input should attract standard copyright protection vesting in that human. Works produced through minimal prompt-based instruction without material creative selection should fall into a new category attracting a shortened protection term of five years, with ownership vesting in the deployer.
A mandatory licensing and compensation framework for training data should be introduced, modelled on the extended collective licensing schemes in Scandinavian copyright law. This would require model developers to obtain licences for training on Indian copyrighted works through a designated collecting society, with proceeds distributed to rights holders.
A labelling obligation should require all commercial AI-generated content to carry a machine-readable declaration of its AI origin. The Copyright Office should be empowered to maintain a public registry of AI-trained models, their declared training datasets, and any licensing arrangements entered.
Conclusion
The death of authorship, as a legal category, is neither inevitable nor desirable. What generative AI demands is not the abandonment of copyright but its disaggregation, a recognition that different creative contributions deserve different rewards, that human creative labour remains qualitatively distinct even when assisted by powerful tools, and that the entities who profit commercially from AI generation must not be permitted to do so on the unpaid creative capital of the communities whose expression built those systems. Authorship will not die. But its boundaries must be redrawn with more precision, and more fairness, than existing law currently provides.
