Irene Susan Joseph
CSI College For Legal Studies
INTRODUCTION
The convergence of large-scale machine learning and internet-scale data acquisition has produced a category of cultural production that the framers of the Copyright Act, 1957 could not have anticipated.1 Take a fairly routine prompt. A researcher in Bengaluru instructs a text-to-image diffusion model to render “a Rajput miniature executed in the manner of Nandyal, recast as a cyberpunk silhouette.” In seconds the system produces a polychromatic output. That output competes for attention in the same marketplace as the human miniature tradition itself.
Algorithmic synthesis of this kind challenges a foundational assumption underlying copyright registries worldwide: that every protected work originates in a singular, identifiable human author.2 In India, the statutory architecture contained in Sections 2(d)(vi), 13, 14, and 17 of the 1957 Act remains tied to the paradigm of the individual literary or artistic creator. GenAI rends that paradigm at its conceptual core.3 The Registrar of Copyrights’s informal withdrawal of the co-authorship claim in respect of the RAGHAV Artificial Intelligence (“Suryast”) registration signalled, in fairly dramatic terms, the absence of any statutory hook upon which non-human authorship could be fastened.
This article argues that the present statutory framework is structurally inadequate to allocate authorship, deter unauthorised ingestion of protected corpora, or license the outputs of computationally derived synthesis. Legislative reform, calibrated to the Indian innovation economy yet cognisant of the European Union’s Directive 2019/790 and the United States’ open-ended fair use doctrine, is constitutionally and economically overdue.
The argument proceeds in four steps. It first situates GenAI within the existing statutory lattice; second, it interrogates the originality threshold inherited from Eastern Book Company v D.B. Modak (2008) 1 SCC 1; third, it subjects training ingestion and output infringement to Section 14(a)(i), Section 52(1)(a), and the ordinary observer test articulated in R.G. Anand v M/S Delux Films AIR 1978 SC 1613; and fourth, it evaluates comparative regimes before advancing reforms tailored to the Indian copyright landscape.
THE LEGAL FRAMEWORK
Indian copyright protection attaches only to works in which “author” denotes a human being who has applied sufficient creative input.4 Section 2(d)(vi) of the Act treats, “in relation to any literary, dramatic or artistic work,” the author as the person who causes the work to be created. Section 13 enumerates the protected categories. Section 14 confers economic rights, including the exclusive right to reproduce under Section 14(a)(i). Section 17 vests ownership in the author, absent contractual assignment.5 Read together, they produce a person-centric regime in which rights arise upon creation and subsist for sixty years after the author’s death.
The statutory frame was conceived in a milieu where every protected work derived from identifiable human cognition. GenAI ruptures this premise. The creative act is bifurcated: there is a distant act of authoring a prompt, and a proximate, computationally executed act of synthesis.6 The Copyright Act presently lacks any deeming fiction treating the computer or its operator as author. The Registrar’s withdrawal notice concerning the co-authorship claim advanced by Ankit Sahni in respect of the RAGHAV Artificial Intelligence (“Suryast”) output is the closest the Indian administrative apparatus has come to a pronouncement on the point.7
The result is a doctrinal vacuum. Drawing it out in explicit terms is the first task of this article. Filling it, through reform, is the second.
III. AUTHORSHIP AND THE ORIGINALITY THRESHOLD
The originality threshold articulated by the Supreme Court in Eastern Book Company v D.B. Modak (2008) 1 SCC 1 governs copyright subsistence in India.8 The Bench accepted that “skill” denotes nontrivial intellectual effort and “judgment” the exercise of discretion, both directed toward the production of a final expression. The Court expressly rejected the older “sweat of the brow” heuristic associated with University of London Press Ltd v University Tutorial Press Ltd (1916) 2 Ch 601, retaining only the distillate of intellectual output that bears the imprint of the author’s personality.9 The proposition is, in substance, defensible. It is also the proposition whose application to machine output is least straightforward.
Two distinct invocations of the threshold need to be distinguished. Where a human conceives a prompt, selects parameters, curates outputs, and renders the final expression through post-production edits, the doctrinal apparatus is ordinarily capable of accommodating that contribution as “skill and judgment” directed toward an identifiable expression. Where a generative system proceeds autonomously — producing outputs that bear no traceable relationship to any human participant in the loop — the threshold is unmet because the statutory author simply does not exist. The Copyright Office’s withdrawal of the Suryast registration exemplifies the administrative instinct to refuse recognition to a non-human source, even though registration is voluntary and constitutes mere prima facie evidence under Section 51.10
A fairer reading of the existing doctrine, and one that has been pressed in recent secondary writing, treats the prompt itself as a sufficient original contribution to vest authorship in the prompter, irrespective of the model’s autonomous interpolation between textual inputs and visual or textual outputs. The reading has real force in narrowly defined workflows where the prompter exercises substantial creative control — say, a researcher iterating through diffusion seed values until a particular compositional outcome is achieved. It is harder to sustain as a universal rule. The Copyright Act demands that the author cause the work to be created, and the architect of a generative pipeline does not, in any ordinary sense, cause the work to be created so much as instruct a probabilistic system whose outputs are necessarily underdetermined by instruction.
There is, then, a continuum that the present statute cannot map with doctrinal precision. Only purposive amendment can resolve it; the case law alone will not.
TRAINING INGESTION, SUBSTANTIAL SIMILARITY, AND INFRINGEMENT
Ingestion as Reproduction
GenAI ingestion implicates two stages of analysis. First, there is the act of copying protected material into training datasets. Second, there is the subsequent generation of outputs that may or may not substantially resemble protected works. Both stages engage Section 14(a)(i), which confers upon the copyright owner the exclusive right to reproduce the work in any material form.
The indiscriminate scraping of textual, visual, and audiovisual corpora for model training constitutes reproduction within the meaning of Section 14(a)(i) where ingested copies are stored in volatile or persistent memory.11 India’s Copyright Act lacks a technology-neutral text and data mining (TDM) exception. The rigid enumeration in Section 52(1)(a) confines fair dealing to research, private study, criticism, review, and reporting of current events, and judicial interpretation confirms the closed-list character of the provision.12
The United States’ open-ended fair use doctrine under 17 U.S.C. § 107 permits commercial transformative analysis to escape liability under certain conditions. The European Union’s Directive 2019/790 harmonised a mandatory TDM exception in Article 4, subject to an opt-out, including for commercial purposes, while leaving Member States free to extend the exception further.13 India’s position lies between these regimes but has been interpreted more narrowly than either. The result is that commercial AI developers operating in or from India face a regime whose precise content is, as the DPIIT working group itself acknowledged, under-determined.
Output Infringement
Whether a generated output infringes depends on the application of the idea-expression dichotomy and the ordinary observer test articulated in R.G. Anand v M/S Delux Films AIR 1978 SC 1613.14 The Bench there held that no infringement arises where the defendant’s work, viewed through the perception of a reasonable and ordinarily prudent lay observer, conveys to that observer a sufficient impression of difference from the plaintiff’s work. Where GenAI outputs reproduce protected expression with sufficient fidelity to trigger the test, infringement will lie — against the deployer, if not against the absent author.
The ordinary observer test was, however, designed for two works of comparable human authorship. Applied to machine output whose expressive content may have been memorised, interpolated, or substantially copied, the test remains serviceable for the liability inquiry. The difficulty is that it presupposes a defendant whose conduct is actionable. Where the work is generated and distributed by an end user with no direct intention to copy, the locus of infringement migrates upstream to the model’s deployer, and ultimately to the provider whose training corpus encoded the protected expression. That migration has no clear doctrinal stopping point under the present statute.
CRITICAL EVALUATION AND COMPARATIVE PERSPECTIVES
The European Union’s Regulation 2024/1689 (the AI Act) introduced, in Article 50, a transparency regime requiring providers and deployers of generative systems to disclose AI-generated content to natural persons.15 The disclosure requirement complements the Directive 2019/790 TDM exception by clarifying the evidentiary basis upon which rights-holders may pursue infringement claims. The United States has, by contrast, pursued a litigation-led approach: multiple suits are presently pending in the District Courts for the Northern District of California concerning the ingestion of copyrighted works by frontier-scale models. A proliferation of sui generis proposals — Japanese, Korean, Singaporean — now sits beside these two anchors. India has, to its credit, begun consultation; the Department for Promotion of Industry and Internal Trade’s Working Paper on Generative AI and Copyright (2024) acknowledged the threefold inadequacy and recommended calibrated reform. Whether the consultation will translate into legislation remains, of course, an open question.
India’s statutory regime suffers three principal inadequacies. The originality threshold derived from Eastern Book Company v D.B. Modak has not been authoritatively extended to the adjudication of prompt authorship. Section 52(1)(a) is doctrinally incapable of accommodating commercial TDM, regardless of transformative purpose, because its enumerated purposes exclude commercial analytics and computational training. Section 17 presumes that there is a human author capable of holding title, a presumption falsified by autonomous generation, and the Act prescribes no rule of default ownership analogous to the work-for-hire doctrine of 17 U.S.C. § 201(b).
It might be objected that judicial interpretation can evolve organically to meet the GenAI challenge. The objection has purchase in jurisdictions whose fair-dealing jurisprudence is open-ended; it has less here. The closed-list construction of Section 52 is a judicial constraint rather than a legislative gap to be filled by interpretive jurisprudence, and the absence of a TDM exception cannot be cured by recourse to Section 52(1)(a) because that provision does not authorise it. Legislative reform is therefore the cleaner path forward, though this article concedes that a prolonged period of judicial experimentation would carry its own pedagogical value. On balance, however, the statutory route is preferable.
CONCLUSION
This article has argued that the Indian copyright framework, conceived for individual human authors, is structurally incapable of allocating authorship, regulating ingestion, or licensing the outputs of generative systems. The originality threshold derived from Eastern Book Company v D.B. Modak cannot, without purposive extension, accommodate prompt authorship or autonomous machine generation. The ordinary observer test of R.G. Anand v Delux Films supplies a workable infringement inquiry but presupposes a defendant whose conduct the Act presently fails to identify. Section 52(1)(a) excludes the very activities on which commercial GenAI depends. Section 17 withholds ownership because there is no author to own.
Three concrete reforms follow. Section 2(d)(vi) should be amended to recognise, by deeming provision, the human prompter as author where that prompter exercises substantial creative control over the expressive content of the output. A tiered TDM exception, modelled on Article 4 of Directive 2019/790, should be introduced into Section 52 — non-commercial analysis permissible unconditionally; commercial analysis subject to an opt-out regime. Statutory licensing of generative outputs should be facilitated through Section 33 Copyright Societies, so that rights-holders may monetise authorised training corpora without fragmenting litigation across individual actions.
Taken together, these reforms would align Indian copyright doctrine with the realities of algorithmic production. They would also, it is hoped, preserve the statutory respect for human creativity that animates the 1957 Act.
BIBLIOGRAPHY
Table of Cases
Eastern Book Company v D.B. Modak (2008) 1 SCC 1.
R.G. Anand v M/S Delux Films AIR 1978 SC 1613.
University of London Press Ltd v University Tutorial Press Ltd (1916) 2 Ch 601.
Table of Legislation
Copyright Act 1957 (India), ss 2(d)(vi), 13, 14(a)(i), 17, 33, 51, 52(1)(a).
Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market, art 4.
Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), art 50.
17 U.S.C. § 107 (Fair Use).
17 U.S.C. § 201(b) (Work made for hire).
Secondary Sources and Online Materials
Bently L, Sherman B, Intellectual Property Law (Oxford University Press, 5th edn, 2023).
Chauhan K, ‘Generative AI, Text and Data Mining (TDM) and the Fair Dealing Doctrine: Examining the New Problem with the Old Regime’ (Journal of Intellectual Property Rights, 2025) <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5096997> (accessed 21 August 2026).
Department for Promotion of Industry and Internal Trade, Working Paper on Generative AI and Copyright (Ministry of Commerce and Industry, 2024) <https://www.ikigailaw.com/article/655/exploring-the-dpiits-working-paper-on-generative-ai-and-copyright> (accessed 21 August 2026).
European Commission, ‘Code of Practice on Transparency of AI-generated Content’ (Digital Strategy, 2025) <https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content> (accessed 21 August 2026).
Indian Copyright Office, Withdrawal Notice to Ankit Sahni concerning RAGHAV Artificial Intelligence (“Suryast”) Registration (Managing IP, 2023) <https://www.managingip.com/article/2a5bqtj8ume32iwlaoy5y/exclusive-indian-copyright-office-issues-withdrawal-notice-to-ai-co-author> (accessed 21 August 2026).
Scaria AG and Jhavar V, ‘Striking the Balance: Adapting Indian Copyright Law for GenAI and Beyond’ (Working Paper, SSRN, 2025) <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5115655> (accessed 21 August 2026).
Verma A, ‘The Copyright Problem with Emerging Generative AI’ (Journal of Intellectual Property Studies, 2023) <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4537389> (accessed 21 August 2026).
FOOTNOTE(S):
1 Copyright Act 1957 (India), s 13.
2 Scaria AG and Jhavar V, ‘Striking the Balance: Adapting Indian Copyright Law for GenAI and Beyond’ (Working Paper, SSRN, 2025) <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5115655> (accessed 21 August 2026).
3 Copyright Act 1957 (India), ss 2(d)(vi), 14, 17.
4 Eastern Book Company v D.B. Modak (2008) 1 SCC 1 [18]–[22].
5 Copyright Act 1957 (India), ss 2(d)(vi), 13, 14(a)(i), 17.
6 Chauhan K, ‘Generative AI, Text and Data Mining (TDM) and the Fair Dealing Doctrine: Examining the New Problem with the Old Regime’ (Journal of Intellectual Property Rights, 2025) <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5096997> (accessed 21 August 2026).
7 Indian Copyright Office, Withdrawal Notice to Ankit Sahni concerning RAGHAV Artificial Intelligence (“Suryast”) Registration (Managing IP, 2023) <https://www.managingip.com/article/2a5bqtj8ume32iwlaoy5y/exclusive-indian-copyright-office-issues-withdrawal-notice-to-ai-co-author> (accessed 21 August 2026).
8 Eastern Book Company v D.B. Modak (2008) 1 SCC 1.
9 University of London Press Ltd v University Tutorial Press Ltd (1916) 2 Ch 601; Eastern Book Company (n 8) [18].
10 Copyright Act 1957 (India), s 51; Withdrawal Notice (n 7).
11 Copyright Act 1957 (India), s 14(a)(i); Verma A, ‘The Copyright Problem with Emerging Generative AI’ (Journal of Intellectual Property Studies, 2023) <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4537389> (accessed 21 August 2026).
12 Copyright Act 1957 (India), s 52(1)(a); Chauhan (n 6) 5–7.
13 Directive (EU) 2019/790, art 4; 17 U.S.C. § 107.
14 R.G. Anand v M/S Delux Films AIR 1978 SC 1613.
15 Regulation (EU) 2024/1689, art 50.
