India’s AI & Copyright Consultation: Why the Proposed Model Misses the Mark
Technology policy often struggles because of how problems are framed. Small definitional choices—about what a technology does, what a legal right protects, or where harm actually occurs—can have outsized regulatory consequences. India’s consultation on generative AI and copyright sits squarely within this global challenge.
The key question is not whether to act, but whether the model being proposed accurately reflects how AI systems work and what copyright law is designed to regulate. Concern and ambition alone are not enough. As currently framed, the proposal risks addressing the wrong problem and, in doing so, creating new ones.
At the heart of the consultation is a proposal for a mandatory, blanket licensing system for AI training: one nation, one licence, one payment. While administratively attractive, this approach rests on a set of assumptions about AI training, copyright relevance, and creative incentives that deserve closer scrutiny.
Below, I outline the core concerns raised by the proposed model and why India would be better served by a clearer, more proportionate framework grounded in legal certainty and technical realism.
AI training is an analytical activity, not an act of cultural consumption
A central assumption running through the consultation is that using copyrighted works to train AI systems necessarily harms human creativity unless creators are compensated for that use. This framing is understandable, but it conflates fundamentally different activities.
Training an AI model is an analytical process. When a system processes large volumes of text, images, or audio, no human is reading the book, watching the film, or listening to the song. The system is not communicating, enjoying, or distributing the expressive content of the work. It is analysing data to identify patterns and relationships.
This distinction matters because copyright law is designed to regulate expressive uses that substitute for, or compete with, the original work in the market. AI training does not operate at that level. It does not function as a replacement for the work itself, nor does it provide audiences with access to the protected expression.
Treating analysis as if it were consumption stretches copyright beyond its traditional role and risks turning it into a general control right over learning itself, something copyright has historically avoided and rightfully so.
Conflating analytical use and expressive reuse creates legal uncertainty
One of the most significant weaknesses of the proposed framework is that it treats all AI training as copyright-relevant by default, triggering payment obligations regardless of whether any expressive harm occurs.
From a policy design perspective, this collapses two distinct stages into one:
- Analytical use: activities such as training, indexing, and data mining, where works are processed but never meaningfully perceived by humans.
- Expressive reuse: situations where protected expression is reproduced or closely mimicked in outputs in a way that competes with the original work.
The second category clearly falls within copyright’s scope. The first does not.
Failing to draw this line explicitly increases uncertainty for all stakeholders. Developers cannot predict when obligations arise. Creators are encouraged to view all AI-related uses as exploitative, even where no expressive substitution takes place. The result is not balance, but friction.
Mandatory blanket licensing introduces complexity, not clarity—and ignores use-case differences
The appeal of a single, government-administered licence is understandable, particularly from an administrative standpoint. In practice, however, a mandatory blanket licensing system raises substantial unresolved questions.
How are individual works to be identified within models trained on billions of data points? How is value attributed to any single work in probabilistic systems whose outputs do not trace back to specific inputs? At what point does “commercialisation” occur: model development, deployment, or downstream use?
These questions point to a deeper issue: the proposed model does not meaningfully differentiate between types of use. Research organisations, cultural heritage institutions, and public-interest projects are treated in the same way as commercial AI developers. Analytical use is treated in the same way as market-facing exploitation.
This lack of differentiation is a policy design flaw. Effective frameworks scale by distinguishing between uses, risks, and contexts, not by flattening them into a single category.
Creating a new payment right is not the same as protecting creativity
The consultation implicitly assumes that mandatory payments for training data are necessary to preserve incentives for human creativity. This assumption deserves closer examination.
Copyright does not guarantee compensation for every indirect or non-expressive use of a work. Its role is to prevent uses that substitute for the original expression in the market. Introducing a statutory remuneration right untethered from demonstrable market harm risks turning copyright into a general levy on technological development.
There is also a structural risk. Blanket royalty systems tend to favour large catalog holders over individual creators, smaller publishers, and regional or minority voices. Without careful calibration, such systems can reinforce existing asymmetries while delivering limited practical benefit to the creators they are meant to protect.
International approaches emphasise differentiation and legal certainty
India is not alone in confronting these questions. Other jurisdictions have approached the issue by carving out analytical uses from copyright while maintaining strong protection at the point of expressive output.
Text-and-data-mining and so-called “non-enjoyment” exceptions reflect a shared policy logic: where content is lawfully accessed and used for analysis rather than communication, permission is not required. Where outputs reproduce protected expression in a market-substituting way, copyright continues to apply.
Crucially, many of these frameworks explicitly distinguish between research and cultural heritage uses and commercial applications. India’s proposed model does not yet make this distinction, increasing the risk of over-regulation and under-inclusiveness at the same time.
Lawful access is a more proportionate safeguard than blanket payment
If the policy goal is to prevent abuse, piracy, or unfair exploitation, there is a more targeted and proportionate tool available: a lawful-access requirement.
Under this approach, AI developers may train only on content they have legitimately obtained through purchase, subscription, user provision, or public availability. This preserves creators’ ability to monetise access to their works without layering an additional compulsory payment onto analytical use.
It also aligns incentives more effectively. Developers are encouraged to rely on legitimate sources. Creators retain control over access. Copyright enforcement remains focused on expressive misuse, where it is most justified.
Why framing choices matter for India’s AI ambitions
India has articulated a clear ambition to be a global leader in AI development, including in multilingual, public-interest, and research-driven applications. Achieving that ambition depends on legal certainty and regulatory proportionality.
Over-regulating the training phase risks favouring large incumbents who can absorb compliance costs, while discouraging domestic startups, researchers, and cultural institutions. The likely outcome is not stronger creative protection, but fewer locally developed AI systems.
A framework that clearly separates learning from copying, analysis from expression, and access from exploitation would better support both India’s creative economy and its AI ecosystem.
A more balanced way forward: clarity, differentiation, proportionality
The consultation rightly recognises the importance of both creativity and innovation. But effective policy requires more than good intentions.
A more balanced approach would:
- Explicitly recognise AI training as a non-expressive, analytical activity
- Should any measure be undertaken at the training level (which we do not believe should be the case), at the very least, differentiate between research and cultural heritage uses and commercial applications
- Target copyright protection against expressive outputs that compete with original works
Copyright works best when it draws clear lines around what it protects. As India refines its approach to AI and copyright, preserving those lines will matter more than expanding them. The real decision is not whether to protect creativity, but how to do so without obscuring the difference between learning and copying.
India’s consultation presents an opportunity to avoid that trap. The real choice is not between creators and AI, but between proportionate policy and regulatory overreach.
Get the article’s core concepts in a unique auditory summary—listen to the AI-generated track below!
This song was created entirely using artificial intelligence tools. If you enjoy this experiment, keep an eye out for each song illustrating our December articles.
Written byCaroline De Cock, LL.M., Head of Research
