Authored By: GOWTHAMI.R
Sathyabama Institute of Science and Technology
Introduction
The rise of generative artificial intelligence has moved the copyright debate in India past its familiar territory of infringement and fair dealing into a far less settled question: when a machine produces a poem, a painting, or a piece of software code, who, if anyone, owns it? Tools such as ChatGPT, Midjourney and India’s own RAGHAV are no longer curiosities confined to technology journals; they are used daily by students, advertisers, journalists and lawyers to generate text and images at a scale no human workforce could match.[1] Indian copyright law, however, was drafted in an era when the assumption that a “work” presupposes a human “author” went largely unquestioned.
This article argues that Indian copyright law should not mechanically extend ownership of AI-generated content either to the AI system or to its developer. Instead, ownership should track the extent of identifiable human creative contribution to the resulting work. The discussion proceeds from the statutory framework and the authorship problem it creates, through the competing ownership claims of the AI system, its developer, the user and the employer, to the emerging Indian case law on the Suryast registration and ANI Media Pvt Ltd v OpenAI OpCo LLC, a brief comparative glance abroad, and finally a proposed human-contribution test with concrete recommendations.
Understanding Generative AI and AI-Generated Content
Generative AI refers to machine-learning systems, typically large neural networks, trained on vast datasets of existing text, images, audio or code so that they can generate new, statistically plausible outputs in response to a prompt.[2] Unlike earlier “computer-assisted” tools such as word processors, where the human operator directs each discrete step, generative systems compress an enormous amount of decision-making into a single statistical process opaque even to their own developers.
For copyright purposes it is useful to distinguish three categories of output: purely AI-generated content, where a short prompt does most of the work; AI-assisted content, where a human supplies substantial original material, such as a photograph, and the AI transforms it, as in the RAGHAV “style-transfer” cases; and AI-influenced content, where a human creates a work independently but consults AI suggestions along the way. The degree of human involvement varies sharply across these categories, and that variation should be decisive for ownership.
III. Indian Copyright Framework Governing AI-Generated Works
The Copyright Act 1957 does not use the phrase “artificial intelligence” anywhere, yet it already contains a provision that anticipates non-human creative processes. Section 2(d)(vi) defines the “author” of “any literary, dramatic, musical or artistic work which is computer-generated” as “the person who causes the work to be created”.[3] This provision, inserted well before generative AI existed, was aimed at outputs such as computer-generated typefaces or database compilations, but its “causes the work to be created” language is broad enough to be pressed into service for AI outputs today.
Section 17 separately governs first ownership, vesting it by default in the author, subject to exceptions for works made in the course of employment.[4] Section 13 confines protection to “original” literary, dramatic, musical and artistic works, sound recordings and cinematograph films, and nowhere extends legal personality to non-human authors.[5] Read together, these provisions create an unresolved tension: section 2(d)(vi) contemplates authorship without direct human execution, while sections 13 and 17 assume the “author” whose originality is assessed, and in whom ownership vests, is a natural or juristic person capable of holding rights.
The originality threshold itself has evolved through judicial interpretation. In Eastern Book Company v D B Modak, the Supreme Court rejected both the low “sweat of the brow” standard and the high American “creative spark” standard, adopting instead a “modicum of creativity” test requiring some, even if minimal, exercise of skill and judgment that is not merely mechanical.[6] This intermediate standard matters for AI-generated works because it leaves room to ask, case by case, whether the human input into a given output crosses the threshold of a “modicum of creativity” or remains merely mechanical prompting.
Authorship and Originality: The Central Legal Problem
The central legal problem is that copyright’s foundational architecture, in India as in most jurisdictions, was built around a presumed identity between the “author” (the person whose intellectual creation the work reflects), the “creator” (the person or process that physically produces the work) and the “owner” (the person in whom the economic rights vest). Generative AI breaks this identity apart for the first time at scale. The computational act of creation is performed by the AI system; the intellectual direction may lie anywhere on a spectrum from a single throwaway prompt to sustained iterative curation by a human; and ownership must nonetheless be assigned to some legal person because Indian law does not recognise the AI system itself as one.
Two doctrinal difficulties follow. First, originality in Indian law has consistently been treated as a human faculty. The “modicum of creativity” standard in Eastern Book Company presupposes an author capable of exercising “skill and judgment”, concepts that describe cognitive choices, not statistical token prediction.[7] Where a system autonomously produces an image from a two-word prompt, it becomes difficult to locate the required “skill and judgment” anywhere except, at most, in the selection of the prompt itself, a thin basis for a strong originality claim.
Second, section 2(d)(vi)’s reference to “the person who causes the work to be created” was never authoritatively tested against a case where several candidates, the developer, the platform operator, and the end user, could each plausibly claim to have “caused” the work. Courts and the Copyright Office have had almost no interpretive guidance on allocating authorship among these competing claimants, a gap the RAGHAV and ANI v OpenAI episodes, discussed in Section VII, have only begun to fill.
Who Owns AI-Generated Content in India?
5.1 Can AI Be an Author?
On existing Indian law, the answer is almost certainly no. Copyright vests in “persons”, and the Act’s definitions of “author” in section 2(d), including the computer-generated limb in section 2(d)(vi), consistently identify a human or juristic person, never the machine itself.[8] Indian commentary has repeatedly noted that authorship “must be a natural person” under settled doctrine, a proposition unaffected by the sophistication of the generative process.[9] Granting authorship to software would also require resolving intractable questions of legal personality and liability that the Act does not address. AI as sole author should therefore remain off the table absent express legislative intervention.
5.2 Can the AI Developer Own the Output?
A stronger, but still incomplete, claim belongs to the developer of the AI system, on the theory that it is the developer who ultimately “causes” outputs to be generated by building and training the model. This argument has intuitive appeal where a developer’s own proprietary dataset and architecture do almost all of the generative work with minimal user input. Yet developers typically disclaim ownership of user-generated outputs in their terms of service, assigning rights (to the extent any exist) to the user instead, precisely because a blanket developer-ownership rule would be commercially unworkable at the scale of millions of daily prompts and would sever ownership from the person who actually selected the creative direction of the output.
5.3 Can the User/Prompter Own the Output?
The user is, on balance, the most defensible default claimant, but only where prompting and curation rise above trivial instruction. Proponents argue that iterative prompts, especially where a user selects among generations, edits outputs, or combines AI output with independently created material, exercise the “skill and judgment” Eastern Book Company requires. Suryast is instructive: Sahni supplied his own photograph, chose the reference style image, and controlled the degree of transfer, well beyond typing a one-line prompt.[10] A user who types “draw a cat in the style of Van Gogh” and accepts the first output has made a functional selection but arguably no original contribution, and rewarding such minimal prompting risks converting copyright into a right over ideas rather than expression.
5.4 Can Employers or Companies Claim Ownership?
Where an employee generates AI content in the course of employment, section 17’s rules should operate exactly as for any other work: ownership vests in the employer absent a contrary agreement.[11] This is the least controversial claim, since it does not require resolving human-AI authorship at all; it simply asks who employed the human whose prompting contribution, if sufficient under Section IX’s test, generated the work. Employers should still build explicit AI-use clauses into contracts, since section 17 offers no certainty where the human contribution is too thin to satisfy originality.
5.5 When Should AI-Generated Content Remain in the Public Domain?
Where no human contribution meets the “modicum of creativity” threshold, and no juristic person can meaningfully be described as the work’s author, the most doctrinally honest outcome is that the output simply attracts no copyright protection at all and enters the public domain immediately on creation. This is not a gap to be regretted; it correctly reflects that copyright is meant to reward and incentivise human intellectual labour, not to manufacture artificial scarcity over outputs that a machine can reproduce indefinitely at negligible cost. A public-domain default for minimally-prompted AI content also guards against strategic over-claiming, where commercial users might otherwise flood the Copyright Office with registrations for content in which their creative contribution is nominal.
Beyond Copyright Infringement: AI Training Data and Copyright
Ownership of outputs is only one half of the emerging problem; the other half, and the one that has actually produced litigation in India, concerns the copyrighted material used to train generative models. Large language models and image generators are trained on datasets scraped from millions of copyrighted books, articles and websites, typically without individual licences. Content owners argue this training itself involves unauthorised reproduction, quite apart from anything the model later outputs. Developers counter that ingesting works to extract statistical patterns is a transformative use falling within fair dealing exceptions such as section 52 of the Copyright Act 1957.[12] Because section 52 lists specific permitted purposes rather than an open-ended fair-use standard, whether “training a model” fits any enumerated category remains genuinely contested, and is precisely what the Delhi High Court was asked in ANI Media v OpenAI.
VII. Emerging Indian Developments: Suryast and ANI v OpenAI
India’s clearest engagement with AI authorship to date is the Suryast episode. Ankit Sahni, an IP lawyer, used RAGHAV, an AI “style transfer” application, to blend his own photograph of a sunset with the style of Van Gogh’s The Starry Night, retaining control over how strongly the style was applied.[13] His first application, naming RAGHAV as sole author, was rejected. His second, naming himself and RAGHAV as co-authors, was accepted in November 2020, reportedly the first instance anywhere of an AI system being registered as a co-author.[14] Within roughly a year, however, the Copyright Office issued a withdrawal notice questioning RAGHAV’s legal status, reflecting its own uncertainty about whether it had acted within its statutory authority. The episode remains unresolved rather than authoritative, but it shows the Office is willing, in principle, to register AI-assisted works where a human retains genuine creative control, while naming the AI itself as “co-author” sits uneasily with the natural-person assumption underlying Indian copyright law.
On the training-data side, the Delhi High Court’s interim ruling in ANI Media Pvt Ltd v OpenAI OpCo LLC, delivered on 24 July 2026, is India’s first substantive judicial pronouncement on generative AI and copyright.[15] Asian News International alleged that OpenAI had trained ChatGPT on its copyrighted news content without a licence, and that ChatGPT sometimes reproduced that content or attributed fabricated statements to it. The Court rejected OpenAI’s jurisdictional objection but, at the interim stage, held that storing and using ANI’s material for training could fall within fair dealing, and declined an injunction, noting ANI had shown no lost revenue and could itself block OpenAI’s crawlers.[16] The ruling is expressly prima facie, and the suit remains pending, but it signals an early judicial willingness to weigh AI’s public benefit against copyright owners’ interests, a balance with obvious implications for how India will eventually treat output ownership too.
VIII. Comparative Perspective: USA, UK and EU
Briefly, three approaches stand out. The United States takes the strictest line: courts and the Copyright Office require human authorship, and refused Sahni’s own US application for Suryast on the ground that it “lacks the human authorship necessary to support a copyright claim”.[17] The United Kingdom is the closest comparator to India: section 9(3) of the Copyright, Designs and Patents Act 1988 already vests authorship of computer-generated works in “the person by whom the arrangements necessary for the creation of the work are undertaken”.[18] The European Union has avoided legislating authorship directly, instead using the AI Act’s transparency obligations on training-data documentation as an indirect lever.[19] India’s section 2(d)(vi) gives it a textual foothold closer to the UK than to the US’s strict rule, strengthening the case for the human-contribution test below rather than wholesale borrowing from either extreme.
Proposed Human Creative Contribution Test
Building on the “modicum of creativity” standard in Eastern Book Company and the existing text of section 2(d)(vi), this article proposes a structured Human Creative Contribution Test, assessed along three factors. First, input originality: did the human supply original expressive material, such as a photograph or draft text, rather than a generic instruction, as Sahni did with his own sunset photograph? Second, directive control: did the human exercise meaningful, iterative control, selecting among outputs or substantially editing the result, rather than accepting a single unmodified output? Third, selection and arrangement: even where individual outputs are not independently original, did the human’s selection or compilation of several involve enough skill and judgment to meet the Eastern Book Company threshold? A work satisfying at least two of these three factors should attract copyright vesting in the human contributor, or their employer under section 17; a work satisfying none should enter the public domain. This keeps the existing statutory language intact and gives the Copyright Office and courts a workable, fact-sensitive framework rather than a rigid binary rule.
Recommendations for India
Three modest reforms would bring clarity without a wholesale rewrite of the Act. First, the Copyright Office should issue formal guidelines clarifying how the Human Creative Contribution Test will be applied to registrations involving AI tools, replacing the ad hoc grant-and-withdrawal approach seen with Suryast. Second, Parliament should consider a narrow amendment to section 2(d)(vi) expressly excluding the AI system itself from qualifying as “author” or “co-author”, leaving the “person who causes the work to be created” language to be read through the contribution test. Third, given the pending outcome of ANI Media v OpenAI, the legislature should watch whether section 52’s enumerated categories need a targeted addition for AI training, rather than leaving courts to stretch categories never meant for this use case.
Conclusion
Indian copyright law is better placed than many commentators assume to absorb generative AI without a wholesale legislative rewrite, because section 2(d)(vi) already contains language capable of directing ownership toward the human who meaningfully shapes an AI-generated work. What is missing is not statutory text but interpretive discipline: a consistent, fact-sensitive standard for separating genuine human creative contribution from trivial prompting. The Suryast registration and withdrawal, and the Delhi High Court’s cautious first engagement with AI training in ANI Media v OpenAI, show a legal system still working out its footing rather than one without a path forward. A Human Creative Contribution Test, grounded in existing doctrine, offers that path: it rewards genuine authorship, denies ownership to machines and to those who merely operate them, and leaves purely machine-generated content where copyright’s incentive rationale suggests it belongs, in the public domain.
Reference(S):
[1]See generally NITI Aayog, ‘National Strategy for Artificial Intelligence’ (2018), discussing AI’s economic role in India, extended in subsequent policy discourse to creative industries.
[2]Ian Goodfellow, Yoshua Bengio and Aaron Courville, Deep Learning (MIT Press 2016) ch 20 on generative models, as subsequently popularised in transformer-based systems.
[3]Copyright Act 1957, s 2(d)(vi).
[4]Copyright Act 1957, s 17.
[5]Copyright Act 1957, s 13.
[6]Eastern Book Company v D B Modak (2008) 1 SCC 1 (SC).
[7]ibid. See also the earlier discussion of originality in University of London Press Ltd v University Tutorial Press Ltd [1916] 2 Ch 601, on which Indian originality jurisprudence has historically drawn.
[8]Copyright Act 1957, s 2(d).
[9]See discussion in commentary on Suryast, noting the settled position that a work’s author must be a natural person, e.g. analyses following the Indian Copyright Office’s 2020 registration and 2021 withdrawal notice concerning RAGHAV.
[10]See Section VII below for full discussion of the Suryast registration.
[11]Copyright Act 1957, s 17(c) (works made in the course of employment under a contract of service).
[12]Copyright Act 1957, s 52.
[13]Registration of Suryast, No A-135120/2020 (Copyright Office, Government of India, November 2020).
[14]See discussion of the Suryast registration and its aftermath in contemporary IP commentary following the November 2020 registration.
[15]ANI Media Pvt Ltd v OpenAI OpCo LLC, IA 45300/2024 in CS(COMM) 1028/2024 (Delhi HC, 24 July 2026, Bansal J).
[16]ibid. See also the Court’s finding that copyright protects expression rather than the underlying facts contained in news reporting.
[17]Thaler v Perlmutter, 130 F 4th 1039 (DC Cir 2025); see also the US Copyright Office’s June 2022 refusal of Sahni’s US application for Suryast.
[18]Copyright, Designs and Patents Act 1988 (UK), s 9(3).
[19]Regulation (EU) 2024/1689 (EU Artificial Intelligence Act), transparency obligations for general-purpose AI model providers.
