Introduction
The development of generative artificial intelligence has changed the creative landscape in ways that were not anticipated by the law at the time. Novels, legal documents, paintings, and musical compositions can now be produced with little human input because to tools like ChatGPT, Midjourney, and Google Gemini. Who owns the copyright in these outputs is a question that has significant legal ramifications.
The Indian Copyright Act, 1957 was enacted at a time when the very concept of a machine authoring a literary or artistic work was confined to science fiction. The Act is built on the assumption that every protected work originates from a human mind. That assumption is now under serious strain. Generative AI does not create by thinking, feeling or intending but generates outputs by detecting statistical patterns in vast training datasets. The output may look creative but the process behind it is computational.
In this paper, we examine whether AI-generated works can attract copyright protection under Indian law. It argues that the current statutory text does not accommodate non-human authorship. It further examines the doctrinal shift brought about by the Supreme Court in Eastern Book Company v DB Modak,[1] which replaced the “sweat of the brow” doctrine with a standard requiring a “modicum of creativity” rooted in the exercise of human skill and judgment. That standard, as this paper contends, cannot be satisfied by a machine.
The paper also examines the February 2026 amendments to the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, which introduced India’s first statutory definition of Synthetically Generated Information (SGI).[2] These amendments do not resolve the copyright question but they reveal the legislature’s growing awareness of the risks that AI-generated content poses. Finally, the paper analyses the observations made by the Supreme Court of India in early 2026 regarding AI-generated fake judgments and compares the Indian framework with the European Union’s AI Act.[3]
The methodology adopted is doctrinal. The paper analyses statutory provisions, judicial decisions and comparative legislative frameworks. The paper’s second section explores the theoretical underpinnings of copyright, following the evolution from the “sweat of the brow” doctrine to the creative standards. The third section examines the pertinent provisions of the Copyright Act, 1957, with a particular emphasis on half 2(d)’s definition of “author.” Current trends in India and outside are surveyed in Section four. Deepfakes and training data issues are discussed in Section five and finally suggestions for reform and conclusions are presented in Section six.
Theoretical foundations
A. The “Sweat of the Brow” Doctrine
For much of the twentieth century, Indian courts applied the “sweat of the brow” doctrine to determine whether a work was entitled to copyright protection. Under this approach, the investment of labour, skill and capital was sufficient to attract protection. Creativity, in the inventive or expressive sense, was not required. A compiler who arranged telephone directories in alphabetical order could claim copyright, provided he had expended sufficient effort.
The doctrine originated in English law. It rested on the intuition that effort deserves reward and that those who invest time and money in producing a work should not see it freely copied by competitors. This was the dominant approach in Commonwealth jurisdictions, including India for a long time.
The practical consequence of the doctrine was significant: it extended copyright to works that contained no meaningful expression of intellectual personality. A list, a compilation, a database and a collection of raw facts could all attract copyright, provided the compiler had worked hard enough. This was the position that the Supreme Court was called upon to revisit in Eastern Book Company.
B. The Modak Standard: Skill, Judgment, and Minimal Creativity
In Eastern Book Company & Ors v DB Modak & Anr, the Supreme Court of India adopted a new standard for originality. The case arose out of a copyright dispute between two publishers of Supreme Court judgments. The appellants, Eastern Book Company, published a law report called Supreme Court Cases (SCC). They claimed copyright not only in their editorial additions but in the copy-edited text of the judgments itself.
The Court, speaking through Justices BN Agrawal and PP Naolekar, held that the standard of originality in India requires more than mere labour. A derivative work must contain the author’s independent skill and judgment and must exhibit more than trivial creativity. Purely mechanical tasks do not meet this threshold no matter how laborious it may be.[4]
Applying this test, the Court found that routine corrections, punctuation changes and the insertion of cross-citations were mechanical enhancements that lacked the requisite creativity. However, the breaking and regrouping of text into coherent paragraphs, the creation of internal paragraph-numbering schemes and analytical notations such as “concurring” or “partly dissenting” reflected genuine intellectual effort and were therefore protected.[5]
The Modak standard is now a settled law in India. It occupies a middle ground between “sweat of the brow” and the higher standard of “novelty” required in patent law. It asks whether the author exercised independent skill and judgment in the creation of the work. This is a human-centric inquiry. A machine does not exercise skill or judgment in any legally meaningful sense. It processes inputs and generates outputs according to mathematical functions. There is no “mind” in the process.
This observation has direct consequences for AI-generated works. If the Modak standard requires independent skill and judgment exercised by an author and if “author” under the Copyright Act means a human being, then AI outputs fall outside the scope of copyright protection as presently understood.
C. Theoretical Perspectives on AI and Authorship
Legal scholars have offered three broad positions on AI authorship. The first position holds that copyright requires a human author and that AI outputs should enter the public domain. This is the position most consistent with the current text of the Indian statute.[6]
The second position argues that the human who designs, trains or operates the AI system should be treated as the author. The creative choices made at the design stage are sufficiently expressive to ground authorship. This position draws support from the treatment of computer-generated works in the UK Copyright, Designs and Patents Act 1988, which vests authorship in “the person by whom the arrangements necessary for the creation of the work are undertaken”.[7]
The third position proposes a new category of sui generis protection for AI outputs, modelled on the database right under European law. This would give economic incentives to developers without requiring the fiction of AI authorship.[8]
None of these positions is currently reflected in Indian law. The Copyright Act, 1957 does not contemplate any of them. This is the doctrinal gap that reform must address.
Statutory framework
A. Section 2(d): The Definition of “Author”
Section 2(d) of the Copyright Act, 1957 defines “author” for different categories of works. The definition reads as follows in its relevant parts: for a literary or dramatic work, the “author” is the person who creates it; for a musical work, the author is the composer; for a work of architecture, the author is the architect; for any other artistic work, the author is the artist.[9]
Notably, Section 2(d)(vi) provides that in relation to a literary, dramatic, musical or artistic work that is computer-generated, the “author” shall be taken to be the person who causes the work to be generated. This provision was inserted by the Copyright (Amendment) Act, 1994. It was introduced primarily to address works produced by rudimentary software programs under clear human direction, not autonomous generative AI systems of the kind that exist today.[10]
Section 2(d)(vi) is frequently cited as evidence that Indian law already accommodates AI authorship. This reading is not convincing. The provision speaks of “the person who causes the work to be generated.” In the case of modern generative AI, identifying a single person who “causes” the output is not straightforward. The prompt author provides an input, but the output is the product of the model’s training, architecture, and stochastic processes. The causal chain is too diffuse to map cleanly onto any single person.
Furthermore, the legislative history of the 1994 amendment does not suggest any intention to extend protection to fully autonomous AI systems. The amendment was a response to concerns about software-generated works of a relatively simple kind. It should not be read as an anticipatory solution to a problem that did not exist in 1994.
B. Section 13: Works in Which Copyright Subsists
Section 13(1) of the Act provides that copyright shall subsist in original literary, dramatic, musical and artistic works. The word “original” is the operative term. As the Supreme Court held in Modak, originality requires the exercise of independent skill and judgment by the author.[11]
If the author of an AI-generated work is the machine itself, the originality requirement cannot be satisfied. A machine does not exercise skill or judgment. It applies algorithms. If the author is taken to be the human operator, the question becomes whether the human’s input, typically a text prompt, constitutes sufficient exercise of skill and judgment to ground copyright in the output. A short, generic prompt almost certainly does not. A more elaborate and creative prompt may come closer, but even then, the link between the human’s expression and the AI’s output is tenuous.
C. Section 17: First Owner of Copyright
Section 17 of the Act provides that the author of a work shall be the first owner of the copyright. For works made in the course of employment, the employer is the first owner. The provision does not contemplate ownership by a machine or by a corporate entity simply by virtue of having trained an AI system.[12]
This creates a practical difficulty. If an AI system produces a commercially valuable output, who owns the economic rights? The developer of the AI? The operator who deployed it? The end user who entered the prompt? The Act provides no clear answer. The absence of a clear ownership rule is a significant gap that discourages investment and creates legal uncertainty.
D. The Position of the Copyright Office
The Indian Copyright Office has not yet issued any formal guidelines on AI-generated works. Registration practice has proceeded on a case-by-case basis. In at least one instance, a registration was granted for an AI-generated work on the basis that a human author was identified. However, the legal basis for such registrations remains contested. Registration does not create copyright but is only prima facie evidence of ownership. A registration granted for an AI-generated work without a legally cognisable human author would not survive a court challenge.
Contemporary trends
A. The February 2026 IT Rules Amendment and SGI
On 10 February 2026, the Ministry of Electronics and Information Technology (MeitY) notified the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, which came into force on 20 February 2026.[13]
The most significant conceptual contribution of the amendment is the introduction of the term “Synthetically Generated Information” (SGI). Rule 2(1)(wa) defines SGI as “information that is artificially or algorithmically created, generated, modified or altered using a computer resource, in a manner that appears reasonably authentic or true.”[14]
This is the first statutory definition of synthetic or AI-generated content in Indian law. The definition is notable for two reasons. First, it is technologically neutral. It covers generative AI, neural networks, and rule-based automation without reference to any specific method. Second, the threshold is perceptual rather than technical. The test is not how the content was made but whether it appears authentic. This deception-focused definition has implications for copyright, even though the amendment is primarily a regulatory instrument addressed to intermediaries.
The amendment imposes several obligations. Significant Social Media Intermediaries (SSMIs) must deploy reasonable technical measures to verify whether uploaded content is SGI. Unlawful SGI, including fabricated documents, non-consensual intimate imagery and synthetic content designed to deceive must be removed. Lawful SGI must carry a visible label and embed permanent metadata or a unique identifier. The takedown window for unlawful SGI is three hours.[15]
The amendment does not resolve the copyright question. It does not state who owns the rights in SGI or whether SGI can attract copyright protection. However, it establishes a regulatory category that Parliament or the courts may in future build upon. The recognition that AI-generated content is legally distinct from human-created content is itself an important step.
B.The Supreme Court and AI-Generated Fake Judgments
In early 2026, the Supreme Court of India confronted the systemic danger posed by AI-generated fake legal citations. The occasion arose in Gummadi Usha Rani v Sure Mallikarjuna Rao, where a trial court in Andhra Pradesh had cited four Supreme Court judgments in a property dispute that did not exist. The Andhra Pradesh High Court had acknowledged the judgments as AI-generated but nevertheless affirmed the trial court’s decision on the merits.[16]
The Supreme Court, in a bench of Justices PS Narasimha and Alok Aradhe, took serious exception to this approach. The Court declared that “a decision based on such non-existent and fake alleged judgments is not an error in the decision-making process. It would be a misconduct and legal consequences shall follow.”[17]
The Court issued notice to the Attorney General for India, the Solicitor General and the Bar Council of India, and appointed Senior Advocate Shyam Divan as amicus curiae to assist with the broader institutional question. The matter reflects a growing awareness that AI “hallucinations” pose unique risks in legal contexts. Language models are trained to produce plausible text, not to verify the accuracy of facts. A fabricated citation looks authentic. It is formatted correctly. It may even contain internally consistent reasoning. But the case never existed.
Subsequently, in Heart and Soul Entertainment Ltd v Deepak Bahry, the Court went further, describing the practice of citing AI-generated fake precedents as a “rampant menace” affecting courts not only in India but across jurisdictions globally.[18]
These observations are directly relevant to the copyright debate. If AI systems can produce fake legal documents that deceive courts and lawyers, the case for treating AI outputs as reliable creative work with full copyright protection is significantly weakened. The epistemic trustworthiness of AI-generated content is a relevant factor in any policy debate about its legal status.
C. Comparison with the EU AI Act
The European Union adopted the Artificial Intelligence Act (Regulation (EU) 2024/1689) on 13 June 2024. It entered into force on 1 August 2024 and will be fully applicable on 2 August 2026.[19]
The EU AI Act does not directly regulate copyright ownership of AI outputs. That question is left to Member State copyright law and, at the EU level, to the Copyright in the Digital Single Market Directive 2019/790. However, the AI Act contains several provisions that interact with copyright.
Article 50 of the AI Act imposes transparency obligations on providers of generative AI systems. AI-generated content must be marked in a machine-readable format. For deepfakes and synthetic audio-visual content, visible labelling is required. These obligations apply from August 2025 for general-purpose AI models and from August 2026 for the full set of transparency rules.[20]
Article 53 requires providers of general-purpose AI (GPAI) models to implement a copyright compliance policy and to publish a sufficiently detailed summary of the content used for training. This is a significant departure from the opacity that has characterised AI training data practices. The GPAI Code of Practice, published in July 2025, elaborates on these obligations. It commits signatories to identify and respect rights reservations when crawling the web and to avoid reproducing copyright-protected content in model outputs.[21]
The EU approach has two features that India lacks. First, it mandates transparency about training data, creating a basis for rights holders to identify and enforce their rights. Second, it creates a technology-neutral definition of AI-generated content, similar in structure to the SGI definition in India’s 2026 amendment. However, the EU Act, like Indian law, does not resolve the authorship question. AI-generated outputs under EU copyright law remain subject to the human authorship requirement, unless a sufficient human creative contribution can be established.
Challenges and infringement
A. The Training Data Problem
Generative AI systems are trained on vast datasets that include text, images, music and code. Much of this material is protected by copyright. The developers of these systems generally have not obtained licences from the copyright owners of the training data. This raises a foundational question of whether the use of copyrighted works to train an AI system constitute infringement or not?
Under the Copyright Act, 1957, the reproduction of a copyright-protected work in any material form, including storage in any medium by electronic means, requires the authorisation of the copyright owner.[22] Training an AI system involves storing, processing and extracting patterns from protected works. On a straightforward reading of the statute, this constitutes reproduction and may therefore amount to infringement.
The Act provides a “fair dealing” exception under Section 52 for purposes such as research, private study, criticism, and review. However, the training of a commercial AI system is difficult to characterise as “research” or “private study” in the sense contemplated by the section. Section 52(1)(a) permits fair dealing for the purpose of research. Whether large-scale commercial training qualifies is uncertain.[23]
This uncertainty contrasts with the position in the European Union, where Articles 3 and 4 of the Copyright in the Digital Single Market Directive 2019/790 provide specific text and data mining (TDM) exceptions. Article 4 allows TDM for any purpose, subject to a rights-holder opt-out mechanism. Indian law has no equivalent. The absence of a TDM exception leaves developers in legal uncertainty and may discourage domestic AI development.
The practical consequence is significant. India is home to a growing AI industry. Without clarity on whether the use of copyrighted material to train AI systems is lawful, developers face the risk of retrospective liability. A reform that introduces a targeted TDM exception, with appropriate safeguards, would bring Indian law into alignment with global standards and encourage domestic innovation.
B. Output Similarity and Reproduction
A related question concerns the outputs of AI systems. Generative models sometimes produce outputs that closely replicate protected works from their training data. This can occur in two ways. First, the model may directly reproduce text or images from the training set, a failure mode sometimes called “memorisation.” Second, the output may be substantially similar to a protected work without being an exact copy.
Under Section 51 of the Copyright Act, infringement occurs when any person does anything that only the copyright owner has the right to do, without the owner’s licence. The question is whether the AI developer, the operator, or the end user is the person who “does” the infringing act. This is a question that Indian courts have not yet answered in the context of generative AI.[24]
The determination of substantial similarity in AI outputs poses additional challenges. Courts have traditionally assessed similarity by comparing the expression of the two works. In the case of AI outputs, the similarity may be stylistic rather than literal. A model trained on a particular author’s works may produce text that mimics that author’s voice without reproducing any specific sentences. Whether stylistic mimicry amounts to infringement is an open question under Indian law.
C. Deepfakes and the 2026 Regulatory Response
Deepfakes pose a distinct set of challenges. A deepfake typically uses AI to superimpose a real person’s likeness onto content they did not create or participate in. The Indian legal system has responded to deepfakes through several avenues: defamation law, personality rights, the Information Technology Act 2000, and the 2026 SGI amendment.
The copyright dimension of deepfakes is less commonly discussed. A deepfake video may incorporate protected audiovisual content from the original footage in which the real person appeared. The creation of the deepfake may therefore involve reproduction of protected material without a licence. At the same time, the deepfake itself, as an AI-generated output, may not attract copyright protection if there is no human author in the relevant sense.
The 2026 amendment addresses deepfakes primarily as a harm to individuals and to the information ecosystem. It requires their removal and mandates labelling. It does not address the copyright status of either the source material incorporated into deepfakes or the deepfake itself as a new work.[25]
Delhi High Court decisions on personality rights provide some guidance. In Anil Kapoor v Simply Life India, the Court granted an omnibus injunction protecting the actor’s likeness, voice, and image from unauthorised use by AI systems.[26] However, the personality rights framework is distinct from copyright. It protects the individual’s identity rather than any specific creative work.
The absence of a coherent legal framework for deepfakes in the copyright context is a significant gap. A work created by superimposing one person’s face onto another’s body is not clearly the original work of any human author. It may infringe the rights of the person whose likeness is used, the rights of the original content creator, or both. The current law addresses these issues through different statutes, creating fragmentation and uncertainty.
D. Liability Allocation
A recurring question in the AI and copyright debate is who should bear liability for infringement. The candidates are the AI developer, the operator who deploys the AI, and the end user who generates the infringing output. Under the current framework, each presents difficulty.
Developer liability raises concerns about chilling innovation. Holding developers liable for all outputs of their systems would create unmanageable risk. Operator liability may be more targeted but depends on the operator’s degree of control over the AI’s outputs. End user liability may be disproportionate if the user did not know and could not have known that the output was infringing.
The 2026 amendment addresses intermediary liability for SGI through a specific carve-out. Rule 2(1B) provides that an intermediary does not lose its safe harbour protection under Section 79 of the Information Technology Act, 2000 merely by removing SGI in compliance with the Rules.[27] This clarifies the position for platforms. It does not resolve the liability question for developers or end users in the copyright context.
Conclusion and suggestions
The existing framework of the Copyright Act, 1957 is not equipped to deal with the challenges posed by AI-generated creativity. The human authorship requirement, implicit in the definition of “author” under Section 2(d) and reinforced by the Modak standard, excludes AI outputs from copyright protection unless a human contribution of sufficient originality can be established. This creates a paradox that commercially significant creative outputs may belong to no one thereby entering the public domain the moment they are generated.
The February 2026 SGI amendment and the Supreme Court’s observations on AI-generated fake judgments together signal that Indian law is beginning to engage seriously with the consequences of generative AI. These developments are welcome but insufficient. A piecemeal regulatory response cannot substitute for principled legislative reform.
This paper makes six specific suggestions for reform.
First, the Copyright Act should be amended to clarify that AI systems cannot be “authors” for the purposes of the Act. This would end the interpretive uncertainty around Section 2(d)(vi) and confirm the human-centric nature of Indian copyright law.
Second, Parliament should consider creating a new category of sui generis protection for AI-generated works. This protection should be narrower in scope and shorter in duration than traditional copyright. It should vest in the person or entity that made the significant creative or financial investment in the generation of the work. This would provide economic incentives for AI development without distorting the authorship framework.
Third, the Act should be amended to introduce a text and data mining exception, modelled on Article 4 of the EU Copyright in the Digital Single Market Directive 2019/790. The exception should permit the use of lawfully accessed material for the purpose of training AI systems, subject to a rights-holder opt-out mechanism and transparency requirements about the datasets used.
Fourth, the Indian Copyright Office should issue guidelines on the registration of AI-generated works. At present, registration practice is inconsistent. Clear guidelines would reduce uncertainty and deter fraudulent registrations.
Fifth, the liability framework for AI-generated infringement should be clarified. The Act should specify the conditions under which AI developers, operators, and end users may be held liable for infringement arising from AI outputs. A graduated liability model, taking into account the degree of control exercised by each party, would be most appropriate.
Sixth, India should engage actively in international negotiations on a global framework for AI and copyright. The existing international instruments, including the Berne Convention and the TRIPS Agreement, are not adapted to AI-generated works. A multilateral solution would be preferable to a fragmented patchwork of national rules.
In conclusion, the arrival of generative AI challenges the foundational assumptions of copyright law. India has a choice. It can allow the courts to resolve these questions incrementally, case by case, over many years. Or it can legislate thoughtfully, creating a framework that is both fair to creators and responsive to the realities of technological change. The second path is preferable. The time for reform is now.
******
Footnotes
[1] Eastern Book Co. & Ors v. DB Modak & Anr, (2008) 1 SCC 1 (Ind.) [hereinafter Eastern Book Co.].
[2] Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, Gazette of India, pt. II sec. 3(i), GSR 120(E) (Feb. 10, 2026) [hereinafter IT Amendment Rules 2026].
[3] Gummadi Usha Rani v. Sure Mallikarjuna Rao, 2026 SCC OnLine SC 341 (Ind.) [hereinafter Gummadi Usha Rani].
[4] Eastern Book Co., supra note 1, para 26 (Naolekar J) (holding that a work must exhibit “independent skill apart from capital and labour” and a “minimal degree of creativity”).
[5] Eastern Book Co., supra note 1, paras 35–41.
[6] Andres Guadamuz, Artificial Intelligence and Copyright, 5(3) WIPO MAG. 14, 16 (2017).
[7] Copyright, Designs and Patents Act 1988, c.48, §9(3) (Eng.).
[8] Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law, ch. 5 (Cambridge Univ. Press 2020).
[9] Copyright Act, No. 14 of 1957, §2(d) (Ind.).
[10] Copyright (Amendment) Act, No. 38 of 1994 (Ind.). Section 2(d)(vi) was inserted by this amendment.
[11] Eastern Book Co., supra note 1, para 26.
[12] Copyright Act, No. 14 of 1957, §17 (Ind.).
[13] IT Amendment Rules 2026, supra note 2. The Rules amend the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, Gazette of India, pt. II sec. 3(ii) (Feb. 25, 2021).
[14] Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, Gazette of India, pt. II sec. 3(ii) (Feb. 25, 2021), r. 2(1)(wa) (as amended by IT Amendment Rules 2026, supra note 2).
[15] IT Amendment Rules 2026, supra note 2, rr. 3(1), 3(3), 4(1A).
[16] Gummadi Usha Rani, supra note 3.
[17]Id. (Narasimha and Aradhe JJ). The Court’s words as reported: “a decision based on such non-existent and fake alleged judgments is not an error in the decision making… it may amount to misconduct and entail legal consequences.”
[18] Heart and Soul Entm’t Ltd v. Deepak Bahry (S. Ct. Ind., decided Mar. 26, 2026) (Bindal and Bishnoi JJ), reported in SC Flags Rampant Menace of AI-Generated Fake Judgments Worldwide, IANS (New Delhi, Mar. 26, 2026).
[19] Regulation (EU) 2024/1689, 2024 OJ (L2024/1689) 1, entered into force Aug. 1, 2024, fully applicable Aug. 2, 2026 [hereinafter EU AI Act].
[20]Id., art. 50.
[21] EU AI Act, supra note 19, art. 53; GPAI Code of Practice, Copyright Chapter, Commitment 1 (AI Office, July 2025).
[22] Copyright Act, No. 14 of 1957, §14(a)(i) (Ind.).
[23]Id., §52(1)(a).
[24] Copyright Act, No. 14 of 1957, §51 (Ind.).
[25] IT Amendment Rules 2026, supra note 2, r. 3(3)(b).
[26] Anil Kapoor v. Simply Life India & Ors, CS(COMM) 652/2023 (Del. High Ct., order dated Sept. 20, 2023) (Ind.).
[27] IT Amendment Rules 2026, supra note 2, r. 2(1B).