Artificial Intelligence and Copyright Law in India

Published On: July 23rd 2026

Authored By: Hardik Gajraj

National Law University, Delhi

Abstract

Artificial intelligence has unsettled Indian copyright law by challenging the traditional assumptions of human authorship, originality, and control over creative expression. Generative AI systems can produce literary, artistic, musical, and audiovisual outputs at scale, yet the Copyright Act, 1957 was drafted for a world in which a human author sat at the centre of every protected work. This article examines two core issues: first, whether AI-generated or AI-assisted outputs can qualify for copyright protection under existing Indian law; and second, whether the use of copyrighted works for training generative AI systems can be justified through current statutory exceptions or fair dealing. The article argues that Indian law contains partial tools, especially Section 2(d)(vi), but lacks the doctrinal specificity needed to resolve disputes over authorship, training data, and liability. It further argues that the present framework risks either under-protecting creators or overburdening innovation unless Parliament adopts a clearer licensing, disclosure, and remuneration model.[1]

Introduction

The rise of generative artificial intelligence has created a legal problem that copyright doctrine can no longer avoid. Systems trained on vast datasets can now produce convincing text, images, music, and code, often in response to a simple prompt, and the outputs may resemble human-created works closely enough to trigger questions about ownership, infringement, and attribution. Indian copyright law, however, remains structured around a human author who creates, owns, assigns, and enforces rights in a recognisable act of expression. That structure is increasingly strained when the expressive result is produced by a model whose operations are statistical, iterative, and only partly directed by a human user.[2]

This issue matters for at least three reasons. First, authors and publishers need certainty about whether AI-assisted outputs can be protected or challenged. Second, developers and deployers of AI systems need clarity on whether training on copyrighted materials requires permission or compensation. Third, the legal system must protect the incentive structure of copyright without freezing innovation in a sector that is now economically significant. India has already begun to confront these problems through policy discussion: DPIIT has published a working paper on generative AI and copyright, and that paper expressly recognises the need to examine both the legality of training on copyrighted works and the copyrightability of AI-generated outputs.[3]

The gap in law is therefore not whether copyright exists in India, but whether the current statute can answer AI questions with doctrinal integrity. Section 2(d)(vi) states that the author of a computer-generated work is the person who causes the work to be created, but it does not explain how that rule works where several actors contribute to an AI pipeline. Nor does Section 52 expressly deal with text-and-data mining or machine learning training. The research question addressed in this article is thus whether Indian copyright law can be coherently applied to generative AI, or whether legislative reform is necessary to regulate authorship, training, and remuneration in a more precise way.[4]

This article argues that Indian law is only partly equipped for the AI era. While the statutory language can support a limited human-centric reading of authorship, it cannot fully resolve the status of training data or the ownership of machine-generated outputs. As a result, the best path forward is not judicial improvisation but targeted statutory reform that clarifies authorship, creates a workable licensing regime for training data, and preserves room for innovation without eroding creator rights.

Literature Review

The literature on AI and copyright in India is still developing, but it already shows a sharp divide between interpretive and reformist approaches. One group of scholars argues that Indian law can accommodate AI by reading Section 2(d)(vi) pragmatically, so that the human who prompts, directs, or meaningfully controls the system is treated as the author. Another group argues that this approach is too strained because it ignores the scale and autonomy of generative models, which may generate expressive outputs without any single person exercising traditional creative control. A third line of writing goes further and proposes a sui generis framework, including compulsory licensing or even a separate remuneration right for training use.[5]

The academic debate has sharpened because the government has now entered the field. DPIIT’s working paper on generative AI and copyright is important not only because it acknowledges the issue, but because it frames the problem as one of policy design rather than one that can be solved entirely through existing doctrine. Commentaries responding to the paper suggest that the absence of a text-and-data mining exception in Indian law is increasingly viewed as a major doctrinal gap. That gap matters because, unlike ordinary infringement disputes, AI training involves repeated copying at scale, often hidden within technical processes that copyright law was never designed to audit directly.[6]

The scholarly debate also reflects a deeper disagreement about copyright’s purpose. Some writers treat copyright as a reward for creative labour and therefore argue that AI outputs lacking human creativity should not receive protection. Others emphasise market function and argue that refusing protection altogether may discourage investment, commercialisation, and responsible use. Still others note that even if outputs themselves remain protectable, the training process may independently interfere with the rights of authors whose works are ingested into datasets without consent. The current literature is therefore rich on diagnosis but less settled on a normative solution that fits Indian statutory design.[7]

That is the gap this article seeks to address. It combines doctrinal analysis with policy critique and comparative insight to argue that Indian law needs a clearer framework for both authorship and training, rather than relying on scattered analogies to old copyright categories.

Legal Framework

The Copyright Act, 1957 is the foundation of Indian copyright law. Section 13 sets out the works in which copyright subsists, including literary, dramatic, musical and artistic works, cinematograph films, and sound recordings. Section 14 defines the bundle of exclusive rights attached to those works. Section 17 establishes the general rule that the author is the first owner, subject to statutory exceptions. Section 19 governs assignment, while Section 57 recognises moral rights.[8]

The most important provision for AI-generated works is Section 2(d)(vi), which treats the author of a computer-generated work as the person who causes the work to be created. This language was written long before modern generative AI, but it remains the statutory hook through which Indian law must currently analyse machine-assisted creativity. The difficulty is that the phrase “causes the work to be created” is under-specified. It can be read to include the prompt-giver, the developer, the platform operator, the commissioner, or some combination of these actors depending on factual control.

Section 52 is equally important, but for the training-data issue rather than output ownership. It lists specific acts that do not amount to infringement, including certain private, educational, and research uses, as well as other limited statutory exceptions. However, the provision does not expressly mention text-and-data mining, machine learning, or model training. The absence of a dedicated AI or TDM exception means that legal analysis must rely on analogy, which creates uncertainty for both rights-holders and developers.

The Constitution matters indirectly because copyright regulation affects expression, innovation, access to knowledge, and the development of the digital economy. While the Constitution does not create copyright, constitutional values shape how copyright exceptions and enforcement should be understood, especially where new technologies affect speech and access. In that sense, AI and copyright is not just a technical issue; it is also a question about how India balances creativity, competition, and access in a digital environment.

Current policy discussion shows that the government is aware of this tension. DPIIT’s committee mandate, as reflected in official materials, includes examining the legality of using copyrighted work in AI training, the copyrightability of works produced by generative AI systems, and international practice. This indicates that the law is currently in a transition phase: the statute remains unchanged, but the policy conversation has already moved toward reform.

Judicial Analysis

Indian courts have not yet produced a settled, AI-specific copyright doctrine, but existing copyright principles remain highly relevant. The first principle is that copyright is generally concerned with original expression, not mere ideas or facts. The second is that originality requires more than mechanical labour; there must be some meaningful creative contribution. The third is that rights and exceptions must be interpreted in light of statutory text rather than technological convenience. Those principles become difficult to apply when an AI system contributes substantially to the expressive form of the output.

Recent litigation has exposed the uncertainty. In 2026, the Delhi High Court reportedly directed the Copyright Office to decide whether an AI system could be treated as the sole author of an artwork, a striking illustration of how unresolved the issue remains in India. The very fact that the issue had to be addressed administratively rather than through settled precedent shows how little direct judicial guidance exists. The court’s caution is understandable because a ruling on AI authorship could have wide consequences for registration practice, ownership claims, and the validity of AI-generated works more generally.[9] [10]

The broader judicial approach in India suggests a human-centric orientation. Copyright protection has traditionally been understood as linked to personal creativity and expressive control, which makes fully autonomous AI outputs hard to fit within existing doctrine. At the same time, courts are likely to be sensitive to the policy consequences of denying protection entirely, especially where AI is used as an assistive tool rather than as an independent creator. This means that the likely judicial path is a distinction between AI-assisted human creativity and wholly autonomous generation.

A further judicial concern lies in minority or underrepresented creators whose works may be used to train models without permission or compensation. Although copyright is not a minority-rights statute in the constitutional sense, it often protects vulnerable creators from market appropriation by larger entities. If AI training systematically consumes the output of smaller creators while concentrating profit in platform owners, courts may need to think more carefully about the distributive implications of copyright enforcement.

Critical Analysis

The central problem in Indian copyright law is ambiguity. Section 2(d)(vi) was drafted to deal with computer-generated works in a relatively simple sense, but today’s generative AI systems are not merely tools that execute fixed commands. They learn patterns from data, produce outputs probabilistically, and may generate content that no single human fully designed in advance. In that environment, the statutory phrase “the person who causes the work to be created” is too vague to serve as a stable legal rule. It gives courts flexibility, but it also invites inconsistent outcomes and strategic litigation.

The deeper issue is that copyright doctrine assumes a chain of human creativity. In the AI context, however, the chain is fragmented. A model developer may create the system, a dataset curator may select the training corpus, a platform may fine-tune or deploy the model, and a user may craft prompts and select outputs. If all of these actors contribute, then the law must decide whether authorship belongs to the person with the greatest creative input, the person with the most direct control, or the person who bears commercial risk. None of those answers is obviously correct, and the statute offers no ranking mechanism.

A second major problem concerns training data. Generative AI depends on large-scale ingestion of copyrighted material, often copied and stored in ways that are invisible to the public. The law must therefore answer whether that copying is infringing at the moment of ingestion, only at the moment of reproduction in output, or not at all if the use is sufficiently transformative or research – oriented. Section 52 does not clearly resolve that issue because it was not designed for industrial-scale data extraction. A broad fair dealing analogy may protect innovation, but it can also become an unfunded compulsory licence in disguise.

A third problem is misuse and enforcement. Even if AI outputs are not identical to underlying works, they may imitate style, structure, or market position in ways that undercut creators without satisfying classic infringement tests. Traditional copyright doctrine is strongest when it can identify substantial similarity or direct copying. It is weaker when the harm lies in model training, style appropriation, or market substitution. That means creators may suffer economically even when they cannot prove doctrinal infringement in a narrow sense.

Comparative experience supports this concern. Jurisdictions that have thought harder about computer-generated works often rely on clearer drafting or specific rules, whereas India still depends on a decades-old statutory phrase. The policy debate around DPIIT’s working paper shows that India may need a hybrid solution: a limited authorship rule for AI-assisted outputs, a transparency obligation for training data, and a remuneration or licensing mechanism for large-scale commercial training. That would be more honest than pretending that ordinary fair dealing can absorb the entire AI problem.

My view is that Indian copyright law should not reward fully autonomous AI outputs in the same way as human creativity unless there is real human creative control. At the same time, it should not leave all AI-generated value outside the law, because that could distort investment and commercial practice. The better solution is a calibrated regime: protection where human originality is substantial, no protection where the machine is the real creator, and a clear compensation model where copyrighted works are used at scale for training. That preserves copyright’s moral logic while acknowledging the economic reality of generative AI.

Comparative Analysis

The United Kingdom has long been the most cited comparator because its copyright legislation contains an explicit concept of computer-generated works. Malaysia and Turkey are also useful comparators because they show how civil-law and mixed approaches can differ in the treatment of authorship and personality. The United States, by contrast, remains strongly attached to human authorship, and EU practice likewise tends to emphasise originality in a way that resists granting copyright to machine-made outputs without meaningful human input.[11] [12]

For India, the comparative lesson is not that one jurisdiction has solved the problem, but that different legal systems have made different policy choices. The UK-style approach is administratively simple but may be too permissive if human involvement is minimal. The US/EU approach is conceptually cleaner but can leave economically valuable outputs in a legal grey zone. India seems to be moving toward a more policy-heavy model that focuses on training-data licensing and remuneration, which may ultimately be more suitable for a large, diverse creative market. [13]

Challenges

The biggest implementation challenge is technical opacity. Rights-holders often cannot know whether their works were used in training, and developers may not be able to disclose all data sources without exposing trade secrets or operational burdens. This makes enforcement difficult even where the substantive legal rule is clear.

Litigation is another challenge because AI disputes are likely to be expensive, expert-heavy, and slow. Smaller creators may struggle to bring claims, while large platforms can absorb legal risk more easily. Public awareness is also limited, especially among artists and writers who may not understand the difference between AI-assisted use and machine-generated substitution.

Resource constraints matter too. If India wants a meaningful AI-copyright regime, it needs not just better statutes but also administrative infrastructure, digitised registration systems, and a body capable of overseeing licensing or dispute resolution. Without that, reform may remain symbolic.

Reforms

India should consider creating an independent body or specialised cell to monitor AI training practices and advise on licensing norms. Such a body could develop standard disclosures, maintain a registry of permitted datasets, and help balance innovation with creator protection.

The law should also provide clear statutory guidelines on AI-generated authorship. A sensible rule would be to protect AI-assisted works only when a human exercises demonstrable creative control, while denying protection to outputs produced autonomously without meaningful human authorship. That would reduce uncertainty at the registration stage and prevent opportunistic claims.

Conclusion

AI has exposed a structural mismatch in Indian copyright law. The current statute can partly address computer-generated works, but it does not clearly answer the questions that now matters most: who the author is, whether training is infringing, and how creators should be compensated.

The future path should be legislative and policy-driven rather than left entirely to courts. India should clarify authorship, regulate training data through transparent licensing, and preserve the incentive structure that copyright is meant to serve. If reform is careful, the law can protect both human creativity and technological innovation without pretending that one automatically fits the other.