by Shivani Singh and Sonal Lalwani
When music moved from CDs and downloads to streaming platforms, copyright law had to confront a new question: what does it mean to use a work when the technology through which it reaches the public has changed?
Generative AI is forcing India to ask a similar question again. But this time, the problem is larger. An AI system may not simply deliver a copyrighted song, article or photograph to a user. It may process enormous volumes of material to train a model, raising difficult questions about access, reproduction, licensing, remuneration and the eventual use of the resulting system.
These questions are not new to copyright laws. Each new wave of technology tests the boundaries of copyright frameworks. This has repeatedly required copyright frameworks to adapt to new ways in which creative works are produced, distributed and monetised. The emergence of online music streaming was one such moment. Generative AI presents a similar challenge, but with a potentially much wider impact because the line between accessing a work and processing it as part of a machine-learning training dataset is considerably harder to define.
This is one of the questions that emerges from State of Intellectual Property in India, a five-year assessment of India’s IP ecosystem prepared by the Advanced Study Institute of Asia, published by Manohar. The report looks beyond the number of patents, trademarks and copyrights being filed, granted and registered to examine the institutions, licensing systems, litigation and commercial structures that determine whether those rights actually create value.
Its findings show an IP ecosystem that is expanding rapidly but also raise a broader question: whether the legal and institutional systems around IP are equipped to manage what happens after a right is created, from ownership and licensing to commercialisation and enforcement.
Copyright is growing, but the system is under pressure
There have been significant changes in the last few years. Copyright applications rose 80.3%, from 24,451 to 44,095. Revenue from copyright increased from Rs 248.81 lakh (2020-21) to Rs 364.76 lakh (2024-25). Of the total copyrights registered during 2024-25, literary works accounted for 65.1% of the total works.
The Copyright Office, like the Patent Office and the Trademarks Office, experienced a major backlog. In 2023-24, a total of 56,260 applications were examined and of these, 45,730 were disposed of, and 38,002 registrations were granted. As of 2024-25, a total of 44,095 applications were filed and of these, 26,767 were disposed of, leaving a total of 17,328 applications pending.
But these numbers do not, by themselves, tell us how effectively intellectual property is being used. A filing, registration or grant is a legal event. It does not necessarily tell us whether an asset has been licensed, commercialised, enforced or converted into economic value.
Music illustrates why these numbers need to be understood alongside the structures through which copyright is exercised. In India, different rights in a musical work or sound recording may be owned and enforced by different entities involved in publishing, performing, recording and broadcasting, including the Indian Performing Right Society Limited, Phonographic Performance Limited and the Indian Singers’ and Musicians’ Rights Association. A digital music service can therefore implicate multiple rights held by different parties.
The courts had to work through this complexity when digital music services emerged. In Tips Industries Ltd. v. Wynk Music Ltd., the Bombay High Court considered whether Section 31D of the Copyright Act, 1957 applied to online music streaming. The Court held that Section 31D did not extend to internet broadcasting or online streaming. The Bombay High Court Division Bench subsequently dealt with the appeal and affirmed the position that Section 31D did not cover internet-based services. The case demonstrated how existing copyright provisions could be tested when technology changed the way creative works were delivered to the public.
New technologies are increasingly creating situations where the law is required to adapt to keep pace with challenges that it was not explicitly designed to address. In the context of generative AI, large language models and other similar technologies are often trained using large data sets, raising novel challenges with respect to copyright. Use of a generative AI technology does not necessarily raise challenges only with respect to the applicability of the copyright provisions. There may be challenges with respect to liability, and how copyrighted works are made available for public use. Further, the Generative AI technology may raise challenges with respect to copyright licensing.
New technologies challenge copyright law and the judiciary to address the impacts of such
technologies on society and the creative economy. The challenge is to what extent and how the existing framework of copyright law, and policies provide a safety net for various economic interests.
AI has made the licensing question bigger
Large language models do not simply publish individual works. They are trained on large datasets containing different kinds of material, which complicates copyright questions beyond whether a user can access a protected work. The Copyright Act, 1957 predates generative AI and large language models, leaving several questions unresolved, including how copyright applies to computational processing and training and what authorisation or compensation, if any, should be required for rights holders.
This is no longer a future scenario. In ANI Media Pvt. Ltd. v. OpenAI OpCo LLC, ANI challenged OpenAI’s use of its copyrighted material in connection with AI training. In July 2026, the Delhi High Court declined to grant ANI an interim injunction. In September 2026, the Division Bench hearing the appeal also declined interim relief at that stage and sought OpenAI’s response. The case remains ongoing and does not yet provide a final judicial answer on AI training and copyright in India.
Alongside the litigation, the government has explored possible regulatory approaches, including blanket or “no-fault” licensing, text and data mining, opt-out mechanisms and extended collective licensing. The discussion paper proposes a blanket licence for training, with a reconciliation fee becoming payable upon commercialisation, alongside a mechanism for administering and distributing the fee.
Music provides a useful starting point, but not a ready-made answer. Digital music required courts and rights holders to determine which rights were engaged, who could license them and how remuneration could reach the relevant rights holders. Just as with other new technologies, the use cases for AI vary and challenge different areas of the law. It’s not as simple as saying training an AI model is the same as streaming a song. Using a model for a specific task raises different legal issues from training that model.
It is important to find a solution that enables the development of lawful and legit technologies without negatively impacting the interests of copyright holders. Considering the negative impact an overly restrictive interpretation of copyright would have on smaller AI companies and the fact that just because a work is published does not mean that it can be copied and used without limitation, it is important to develop a system where the interests of all the stakeholders can be protected, and the technology can be developed within the limits of the law.
That makes the question broader than whether AI training constitutes infringement. It also raises practical questions about how a licensing framework could work: who can grant permission, whether permissions can be obtained at scale, how creators can express their preferences, how remuneration should be calculated and who should administer it. India’s existing experience with collective rights management may offer part of the answer, but AI will test those mechanisms at a potentially unprecedented scale.
India’s next IP challenge is what happens after creation
The report’s wider findings show that this is not only a copyright question. Across patents, trademarks, designs and copyright, the central challenge is increasingly what happens after a right is created. The same challenge is visible in AI. India is investing in compute, datasets, foundation models, applications and skills through initiatives such as the IndiaAI Mission, while questions of ownership and commercialisation are becoming more important.
India does not have a framework establishing uniform rules for ownership and commercialisation of IP generated through federally funded research unlike the Bayh- Dole Act in United States law which lets universities, non-profits, and small businesses own and commercialise inventions made with federal research money. Other Asian countries like China [with Science and Technology (S&T) Law Revision (2007)] and South Korea [with Korea Technology Transfer Promotion Act (2000)] also have similar laws which support commercialisation of R&D done by publicly funded research organisations or academic institutions.
For India, the comparison points to the importance of a clearer framework governing who owns IP arising from publicly funded research and how that IP can move from research institutions to the market. A framework that connects ownership with technology transfer, licensing and commercialisation could provide greater clarity for universities, public research institutions and private partners, while ensuring that publicly funded research has a pathway to wider application. As AI becomes increasingly dependent on public investment and publicly supported datasets, these questions will become harder to separate from the broader IP ecosystem.
This connects the report’s wider findings. India is no longer simply trying to generate more IP. The numbers show that activity is already increasing. The challenge now is to build mechanisms that allow IP to move from creation to ownership, licensing, commercialisation and enforcement.
The transition from physical media to streaming showed that copyright does not disappear when technology changes. It has to adapt to new forms of use. Generative AI is asking India to do that again, but across an ecosystem that extends beyond music and entertainment into journalism, publishing, datasets, software, research and public digital infrastructure.
The question is therefore bigger than whether an AI model can train on a particular newspaper, song, photograph or book. It is whether India can build an IP infrastructure that allows technological innovation to draw on human creativity while ensuring that the rights and economic interests attached to that creativity remain meaningful.
Note: The report can be found as a book on major ecommerce sites under the name “State of Intellectual Property in India: A Flagship Report on Five Year of a Granular Data, Full Case-Law Synthesis, International Benchmarking, and an Institutional Landscape”]
About the Authors: Shivani Singh is Centre Lead, Centre for Law and Emerging Technologies at the Advanced Study Institute of Asia. Her work sits at the intersection of law, intellectual property, and emerging technologies. She writes and convenes policy dialogues on technology and regulation.
Sonal Lalwani is Centre Lead, Centre for Competition and Economic Resilience at the Advanced Study Institute of Asia. Her work focuses on technology, innovation, competition, and intellectual property. She holds an LL.M. in IP Rights and Technology Law from Jindal Global Law School and has published on diverse aspects of intellectual property.
The views expressed are personal.
Also Read:
