Introduction

The rise of Generative Artificial Intelligence (“GenAI”) has forced copyright law into unfamiliar territory. Courts across jurisdictions are now grappling with a question that existing copyright statutes are ill-equipped to answer. Can the large-scale copying and storage of copyrighted works for training AI models be justified under copyright exceptions that were originally intended for human activities such as learning and research? In India, this question first came before the Delhi High Court in the high-profile case of ANI Media Pvt. Ltd. v. OpenAI OpCo LLC. In an order delivered on 24 July this year, the Court declined to grant an interim injunction to ANI and held, prima facie, that reproducing and storing copyrighted works to train large language models amounts to fair dealing for “private or personal use, including research” under section 52(1)(a)(i) of the Copyright Act, 1957.

The Court undertook a detailed analysis of the law on fair dealing. It examined the meaning of “research”, the role of transformative use, and the factors relevant to assessing fairness in the context of AI training. In doing so, it became the first Indian to engage directly with the copyright implications of GenAI, albeit at the interim stage. The order is also significant given that, in India, most intellectual property suits are settled before they reach the trial stage. The grant or refusal of interim relief can have a significant bearing on settlement, as the party that secures such relief often gains greater leverage in negotiations.

The order is too recent to have generated an extensive body of academic commentary. The most comprehensive critique to date is Pascal Hetzscholdt’s AI-assisted analysis published on Substack, which argues, among other things, that the Court conflates secrecy with private use, overlooks established considerations such as the amount copied and necessity, and wrongly allows transformative use to “manufacture a missing statutory purpose” rather than merely informing the fairness analysis once a statutory purpose has been identified. We do not revisit those criticisms here. A different perspective has been offered by Vishno Sudheendra on SpicyIP, who welcomes the order as a principled resistance to copyright maximalism, particularly for treating section 52 as a charter of “user rights” deserving a broad and liberal interpretation. We, however, take a contrary view.

We advance a more fundamental critique about the order. Our claim is not that the Court necessarily reached the wrong conclusion, but that it reached a defensible conclusion by an indefensible route. There is nothing objectionable about courts adapting settled legal principles to new technologies. Our concern lies in the manner in which the Delhi High Court has done so. We develop that argument in three parts, each addressing a distinct aspect of the Court’s reasoning. First, we argue that the Court reads “research” as independent of the “private or personal use” to which Parliament has tied it since 1994, departs from Jiwan Publishing House, and relies on a Canadian decision interpreting a materially different statutory provision. Second, we argue that section 52 is not the kind of provision courts may enlarge through interpretation (including through updating construction) and that, in doing so, the Court sidesteps the coordinate Bench’s holding in Chintamani Rao. Third, we argue that the Court’s fairness analysis omits the amount and substantiality of what was taken, despite relying on authorities that treat it as part of the inquiry, and further imports transformative-use reasoning from a different statutory limb without explaining why it applies.

Taken together, these arguments show that the Court repeatedly cited authority without engaging with the propositions that constrained its reasoning. The result is an interpretation of section 52 that expands the provision beyond what its text permits.

At the outset, we clarify that we do not argue that GenAI training must necessarily fall outside copyright exceptions, nor that courts should refrain from engaging with this question until Parliament intervenes. Whether, and on what terms, copyright law should permit the use of protected works to train artificial intelligence is ultimately a policy question on which reasonable disagreement is possible. Our narrower claim is that where the Copyright Act is silent on the use of protected works for AI training, any expansion of the fair dealing exception must rest on a persuasive reading of the statutory text. In our view, the Court’s reasoning falls short of that standard.

Severing Research from Private Use

Section 52(1)(a)(i) provides a fair dealing exception to copyright infringement for private or personal use, including research. It operates as a defence once infringement has been established.

The Supreme Court has recognised that the meaning of “include” depends on the language, context, and object of the particular provision. In Peerless General Finance, for instance, the Court noted that “include” may enlarge the ordinary meaning of a term, but may also operate in a restrictive sense where it is used in the formulation “mean and include”, depending on the statutory context. Here, the structure and history of section 52(1)(a)(i) support the latter reading.  In 1994, an earlier formulation in the statute of “research or private study” was replaced by an amendment with “private use, including research”, and another 2012 amendment subsequently changed this to “private or personal use, including research”. Read against that legislative history, “including research” is best understood as identifying research as a form of the private or personal use protected by the provision, rather than restoring research as an independent purpose. Otherwise, the 1994 amendment would have little to no substantive effect. Moreover, treating research as an independent ground through an expansive reading of ‘includes’ would also allow for commercial, non-private research to be protected. As we show below, this is not how the provision has been interpreted in India.

It is then worth asking what then makes a use “private or personal”. The intuitive answer, that a use is private simply because it is hidden from public view, cannot be right. Confidentiality can attach to any activity, including one that is plainly commercial. A pharmaceutical company’s research programme, conducted within a secure laboratory, is no less inaccessible to outsiders than a student’s photocopying of a few pages to prepare for an examination. Yet only the latter would ordinarily be described as “private” in the sense contemplated by any copyright exception.

The Delhi High Court has already drawn much the same distinction in Jiwan Publishing House. A commercial publisher had reproduced the CBSE’s past examination papers without a licence and argued that its copying fell within section 52(1)(a)(i), since the papers were ultimately intended for students engaged in research and private study. The Court rejected the argument, holding that the exception protects acts of research or private study themselves, not commercial reproduction undertaken with the expectation that the resulting copies will eventually be used for those purposes. What disqualified the publisher was the capacity in which it copied, namely for commercial exploitation, rather than any failure to keep its operations confidential. Its printing and distribution activities were, in an ordinary sense, entirely internal to the business. Yet that did not make its copying “private.”  The same view has been expressed in commentary. J. Sai Deepak argues that it would “defeat the intent of the provision” to treat copying as private merely because “the purchaser of a photocopied material would ultimately put the material to private and personal use.” On that reasoning, he observes, “every copyrighted material could be photocopied without the consent of the owner.”

It is not as though the ANI Court entirely overlooked this authority. It cited Jiwan Publishing, reproduced the relevant passage, and sought to distinguish it (ANI, ¶¶ 186–187). It read Jiwan Publishing as merely denying the defence to a publisher whose books are ultimately used by third parties, such as students, for research or private study, whereas OpenAI stores the works for its own training rather than distributing them onwards.

This differentiation does not withstand scrutiny. In Jiwan Publishing, the publisher lost the defence because it copied the works for commercial exploitation. The fact that students later used those books for research was immaterial. The Court itself stated that the defence “would not be available to such a publisher though the book published by him may be used or be meant for use in research or private study” (ANI, ¶ 186, quoting Jiwan Publishing House, ¶ 21). The downstream research use by students could not bring the copying within section 52. OpenAI stands in no different position. It reproduces ANI’s works as part of its commercial operations, and the fact that “research” takes place within its training process cannot do what the students’ research could not do in Jiwan Publishing. The confidentiality of that process adds nothing. If anything, the publisher’s books were ultimately used by others for study, whereas OpenAI is itself the commercial beneficiary of its copying.

The Court’s reliance on comparative law is equally problematic. To support its conclusion that “research” includes commercial activity, it relied on the Supreme Court of Canada’s decision in CCH Canadian Ltd. v. Law Society of Upper Canada, which held that “research” should be given “a large and liberal interpretation” and is “not limited to non-commercial or private contexts” (ANI, ¶ 180). The Court justified this reliance on the footing that section 29 of the Canadian Copyright Act is pari materia with section 52(1)(a) of the Indian Copyright Act.

However, this premise is incorrect. Section 29 exempts fair dealing for the purposes of research, private study, education, parody, and satire. Research and private study are thus separate and independent grounds. Section 52(1)(a), by contrast, has, since the 1994 amendment, protected fair dealing only for “private or personal use, including research.” CCH could hold that research need not be private precisely because the Canadian provision does not make privacy a subset of research. Section 52(1)(a), however, does.

Furthermore, CCH did not hold that commercial research is automatically fair. It held only that the commercial character of a dealing does not, by itself, exclude it from the scope of “research.” Whether the dealing is ultimately protected depends on a separate and searching assessment of fairness. The Supreme Court of Canada made this explicit in Bell Canada. There, online previews of musical works were held capable of facilitating consumers’ research, yet the dealing was found fair only after the Court had weighed a series of specific factors: that the previews were short, streamed rather than retained, automatically deleted after playback, of lower audio quality than the works purchased, non-substitutive and accompanied by technical safeguards. The breadth of the Canadian gateway to “research” is therefore matched by a correspondingly rigorous fairness inquiry. It considers the amount taken, the availability of alternatives, the necessity of the dealing, and its effect on the market.

This is not to say that the ANI court completely dispensed with the requirement of fairness. It formulated and applied a fairness test of its own. The difficulty is that the test it adopted bears little resemblance to the one that accompanies the Canadian conception of “research” on which it relied. In this test, the Court asked (i) whether the use was confined to training, (ii) whether it competed with or caused market prejudice to ANI, and (iii) whether it served the public interest. It did not consider the factors that perform the disciplining function in Canadian law, most notably the amount taken and the availability of non-infringing alternatives. We examine these omissions in more detail in Part II.

To reiterate, our difficulty is that the Court’s interpretation severs “research” from the private or personal use to which Parliament has expressly tied it since 1994. Even if we assume that interpretation to be textually plausible, a more fundamental question still remains. Is section 52 even the kind of provision whose boundaries may be expanded through judicial interpretation, or is that a choice for Parliament alone? The next section considers that question.

Chintamani Rao and the Limits of Judicial Enlargement

Section 52 begins with “the following acts shall not constitute infringement of copyright, namely…” This choice of words is significant since it indicates that the provision is exhaustive. Unlike expressions such as “including” or “such as”, which introduce an illustrative rather than exhaustive list, the phrase “the following” confines the exception to whatever the Parliament has chosen to enumerate. Courts may interpret the scope of those acts, but they cannot enlarge the list by recognising uses that Parliament did not include.

The Delhi High Court itself recognised this principle in Chintamani Rao. The defendant there argued that the fair dealing exception should extend to cinematograph films and sound recordings, even though those categories were excluded from the text as it then stood. In support, it relied on Clause 31 of the Copyright (Amendment) Bill, 2010, which proposed to expand the exception to cover “any work.” Justice Sanghi rejected the argument. He reasoned that a bill that had not yet been enacted could not determine the meaning of the existing statute, nor was it the Court’s role to anticipate Parliament by reading a proposed amendment into the law before its enactment. In his view, the provision, as it then stood, was clear. It was not a court’s task to legislate under the guise of interpretation.

This was the precedent the ANI Court most needed to confront, yet it did not. While the judgment cites the Single Judge’s decision in Chintamani Rao, it relies on that decision only for its observations on proportionality and the four-factor fairness inquiry (ANI, ¶¶ 136.13, 232). The Court engaged closely with the decision and drew upon it for narrower propositions. It did not, however, engage with its holding that the purposes in section 52 were deliberately chosen by Parliament and that courts cannot enlarge those under the guise of interpretation.

The two objections above concerned whether the Court was entitled to interpret section 52(1)(a)(i) as it did in the first place, before reaching the question of fairness. A possible response to our second argument is that Chintamani Rao is distinguishable since that case concerned extending fair dealing to an entirely new class of works, whereas the ANI Court merely adopted a broader reading of the existing statutory purpose of “research” through the doctrine of updating construction. In Part II, we explain why this response is unpersuasive and develop our third argument that, even assuming the Court was correct to read Section 52 expansively, its fairness inquiry was inadequate. We argue that the inquiry is deficient both because it omits consideration of the amount and substantiality of what was taken and because it imports the transformative use doctrine from a different statutory limb without adequate explanation.

The authors are students of the Batch of 2028, B.A.,LL.B. (Hons.) at the National Law School of India University, Bangalore.