In this submission to the SpicyIP–Jhana Blogpost Writing Competition 2025, Avani Marudwar argues for decentering fair dealing analysis in Indian Copyright law in exchange for treating fair dealing as one of the components of the copyright system as a whole. She further argues that the GenAI copyright debate presents an opportunity to facilitate a structural relook at Indian Copyright law along these lines.
Avani Marudwar is a 5th-year law student at Nirma University and an alumnus of the SpicyIP Summer School 2025. She is a drummer and an IP enthusiast. After learning to stop romanticizing the author, she tries to look at most issues as an artist and a lawyer. A small disclaimer for this post: after struggling to reduce the size of the first draft, the author has used perplexity to reframe the structure and bring it within the word limit.
Dealing Away Fair Dealing: The Curious Case of Fair Dealing and Its Infringement
By Avani Marudwar
The Department of Intellectual Property Protection (DIIPT) has proposed a blanket licensing hybrid model for AI training coupled with a fixed remuneration to copyright holders whose works have been used for training, a development that may seem beneficial on its surface. Yet it hints at the uneven ways in which we conceptualize copyright itself. Academic discourse on the AI disruption has several nuanced branches. I wish to focus on India’s inheritance of Generative AI-Copyright issues. GenAI and all the wonderful complications it brings is being discussed in India within the conceptual boundaries set by the Global North, where its friction with copyrighted works was bred domestically and adjudicated upon without a template. American litigation is driving the Fair Use Train, and India seems to be buying a ‘modified’ ticket. Through this piece, I wish to highlight the preoccupation with copyright as a boundary for the anxieties GenAI brings, which is a trap with fair use being this trap’s biggest bait. This focus on fair use, especially by the Indian judiciary, puts copyright’s foundational architecture at stake. I wish to argue that the Copyright GenAI debate presents an opportunity to reexamine Copyright law’s structural character and to centre other access-enabling principles embedded in Copyright law, rather than being bogged down by debates around fair dealing. Thus, as Pamela Samuelson highlights, fair dealing is meant to be a component of a balanced system, not its entire character.
The AI Problem and the Proposed Solution
The AI copyright question stems at the input and output levels. The first is whether using copyrighted works to train generative AI models constitutes fair use, as Meta did with shadow libraries. The second concerns when AI systems unauthorizedly reproduce copyrighted works in their final outputs. The proposed blanket license attempts to address the first by permitting the use of copyrighted material for training in exchange for fixed royalties to creators. On paper, this balances incentive and access. However, as Swaraj Barooah has pointed out, the proposed exception does not incentivise creative work at all. Instead, it is a step towards the enshittification of the public sphere (Note: This is not to dismiss the difference in exception, exemption, limitation, etc. Pardon me, as I use ‘exception’ as a catch-all term for convenience).
Creators would likely respond to this blanket license by implementing paywalls, subscription models, and anti-scraping technologies to protect their works from AI training. As a result, access would suffer and be bartered away for personal data. The alternatives remain locked behind technological barriers.
The status quo is that copyright holders must prove that their work is being infringed, according to Section 51 of the Copyright Act, by opaque (black box) AI models, while the alleged infringer holds all the training data. The evidentiary burden, thus, falls on the injured party with an unignorable power dynamic. The blanket license proposal flips this burden for non-payment cases and imposes transparency obligations on AI developers, one could say, a necessary intervention.
The DIIPT working paper has a dedicated section on fair dealing, where it concludes that training AI systems gives rise to copyright infringement and that fair use/ fair dealing may be applicable, but it makes no attempt to resolve this question. But herein lies the crux: Why does fair dealing need such constant rescue operations? Why must we continuously propose exceptions, remuneration schemes, and transparency mandates just to keep it functional?
The Root Problem: Fair Dealing Overplayed, Copyright Mischaracterised
The answer lies in how copyright has been conceptualised. Copyright is not simply an author’s right of exclusion, temporarily limited by exceptions. As Prof. Balganesh has laid out, it is dual in character; it aims to balance upstream creativity (the author’s incentive) and downstream creativity (the public’s access to knowledge for further creation and dissemination). This balance is not achieved through fair dealing alone; it is embedded in copyright’s entire statutory framework. As Lokesh Vyas argues, increasing reliance on fair dealing creates ‘faith-based understanding’ that masks copyright’s structural clarity, reducing it to author-public balancing rather than foundational limits like originality and merger.
Consider the recent ANI v. OpenAI case. The court’s framing focused predominantly on whether OpenAI’s use constituted fair dealing and whether it was transformative, commercial, or competitive. The first two issues decided were in relation to whether the storage or use of the ANI’s material would be an infringement. (Again, supplementing the third issue of fair use, as it is an exception after all) With only the first issue mentioning that ANI’s copyrighted content is in the nature of news. Why not a separate issue to adequately address whether ANI’s news articles and footage possessed the originality and subject matter specificity required for copyright protection in the first place? (More on this here). News reporting has limited ways of expressing facts. There are only so many ways to describe the happening of an event. This suggests that much news content may fall under the merger doctrine, which eliminates copyright protection where expression and idea merge. Had the court grounded its analysis in subject matter limitations rather than jumping to infringement and fair dealing, the discussion could have involved compensated limitations like licensing agreements benefiting both parties, becoming a remedy. Maybe preempting subsequent litigation that ANI has pursued against many YouTubers. (Read more on this here and here)
The consequence of this fair-dealing-centric approach is that copyright’s character becomes hollow. It is reduced to a presumed right constantly under siege by an exception. The full architecture, originality requirements, subject matter limitations, term restrictions, and yes, fair dealing disappear. When this happens, the author’s domain asserts itself unchecked. Access suffers. The public domain contracts. Given the above premise, it is important to note that fair use is the hinge that the current information politics rests on.
When Fair Dealing Rights Are Infringed
Yet even when fair dealing rights are properly understood, they face formidable obstacles. Fair-dealing determinations are necessarily costly ex post judicial pronouncements that are usually beyond the reach of most fair dealers. These costs are not conceptual but structural, built into the legal and technological architecture. Take, for example, Sections 65A and 65B of the Copyright Act, which address Technological Protection Measures (TPMs). These sections prevent circumvention of digital locks. The barrier is procedural. Section 65A’s vagueness regarding “intent to infringe” creates ambiguity about what constitutes unlawful circumvention.
When law speaks of intention, consent, negligence, or mens rea; for corporations, states, or individuals. It is not real human minds, but rather semantic constructs and role-bundles generated by legal discourse to sustain communication, including for natural persons.
In the TPM context, this constructed “intent” becomes a weapon, intermediaries can allege it preemptively against fair users, forcing them to bear litigation costs and privacy risks under the proviso to Section 65A(2)(a), which mandates recording circumvention and user identification. This vagueness tilts the power dynamic sharply. Large information intermediaries can bring preemptive legal action against users attempting to circumvent TPMs, regardless of the likelihood of success.
The fair user must internalise substantial legal costs to obtain sanctions permitting circumvention. Fair use as a right ideally guarantees access for the scenarios mentioned in Section 52 of the Copyright Act. It is a statutory assurance that acts of infringement for the generation of knowledge are allowed. Hence, this right must be strong and has the added burden of being explicitly clear to not disincentivize authors.
The result is this: fair dealing rights are technically infringed not through copyright enforcement but through technological and legal barriers erected, as well as due to associated enforcement costs.
Restoring Copyright’s Integrity
Every claim of copyright infringement is an encroachment on the public domain. The preoccupation with fair dealing obscures wider issues. But not every copyright question should be decided on fair dealing grounds. Some disputes concern whether copyright subsists at all (originality, subject matter) as a threshold issue, even before the question of infringement arises. Other concerns include whether the legal framework governing enforcement, particularly TPMs and their exemptions, properly enable fair dealing rather than preventing it.
Only when these foundational elements are properly established can fair dealing operate as intended. Not as a desperate exception fighting for survival, but as one component of a balanced copyright system serving both the author’s incentive and the public’s access to knowledge.
The remedy for technology-enabled IP restrictions is not to fight harder over the fair dealing exception or a blanket license. It is to fight for the integrity of copyright law itself, by clarifying which disputes concern subject matter, which concern fair dealing, and which concern the legal and technological barriers that prevent fair dealing’s exercise. When copyright is understood as a whole, not merely as an exception under constant threat, we can properly address the AI question not as a crisis requiring broad blanket emergency exceptions, but as one instance of a recurring structural problem deserving solutions.
Fair dealing will remain under siege so long as copyright is conceptualised only through its exception. The AI moment offers an opportunity to reframe the entire conversation.
I would like to thank Lokesh Vyas for his invaluable guidance and his comments on all my revisions. I’d also like to thank the SpicyIP team for their comments in refining my submission.