Introduction

As we enter 2025, disruption on a scale not seen in decades has begun to take place within India’s Intellectual Property (IP) landscape due to the rapid growth of Generative AI (GenAI). GenAI has evolved from being just another cool technology to a key legal opponent. In 2023/2024, the world had many theories on how AI would interact with Copyright Law. However, 2025 will be known as the year that India faced these challenges head-on.

The core of this tension revolves around one basic question: Is the training of LLMs (large language models) by using copyright-protected works a violation of those copyrights? Or does the training of these LLMs fall within the exception for “fair dealing” which exists in the Copyright Act of 1957? Presently, India is at a pivotal moment. There are ongoing legal proceedings related to the landmark case ANI Media Pvt Ltd v OpenAI in the Delhi High Court, as well as a committee established by the government that is trying to provide a policy framework on this issue. In this blog, I will discuss the key legal arguments related to the above-mentioned matter and provide an analysis of the lack of established national policies, as well as propose a potential way forward for the development of Indian IP Laws.

The Spark: ANI Media v. OpenAI

Indian IP watchers were keenly interested in watching the progression of these types of cases, like The New York Times v. OpenAI in the USA, for several years before they arrived on Indian soil toward the end of 2024 and into 2025. The largest news agency in India, Asian News International (ANI), sued OpenAI for allegedly using ANI’s copyright-protected news articles to train ChatGPT without permission.

While it is clear that this is not just another commercial dispute, this case will help establish whether the Global South has rights to enforce copyright protection. ANI has made clear that its main argument is that OpenAI has used a substantial amount of proprietary information to generate outputs that are in competition with what the originators produced. When ChatGPT creates a summary of an ANI article, or when it “hallucinates”, or falsely attributes to ANI quotes that are not accurate, it violates ANI’s moral and economic rights.

OpenAI has defended its position based on the technical way the model is trained, claiming that while they do not “copy” the text in the traditional way, they “learn” the statistical probabilities from the text. However, according to the Copyright Act, 1957, Section 14, only the owner of the copyright has the exclusive ability to reproduce a work. The Delhi High Court’s comments in 2025 suggested that the judiciary is keenly aware of the reputational and economic harm potential, moving beyond technical defences to look at market impact.

Legal Quagmire: Section 52 and “Fair Dealing”

In India, the defence for AI companies is based on section 52 of the Copyright Act, which provides a limited list of exceptions where infringement does not occur. In contrast to the United States, which has a very wide fair use agreement established by judges (and evaluated through a four-part test), section 52 is much narrower in its fair dealing provisions.

AI training without licensing is legal in India only where AI training falls under one (or more) of the specific fair dealing exemptions provided for in Section 52, of which only two are worth noting for AI training:

1. Section 52(1)(a) – private use or personal use (including research); OR
2. Section 52(1)(p) – reproduction of works for use by a court.

However, neither exception fits the scale at which LLMs are developed commercially, and the definition of “research” as applied to a for-profit corporation like OpenAI or Google is an area of disagreement. Additionally, historically, the courts in India have been reluctant to interpret section 52 in a way that expands it beyond its intended form. Therefore, the process of creating copies of works in LLM training (i.e., copies stored in a computer’s temporary memory) constitutes infringement unless it falls under one of the statutory exemptions as provided for in section 52.

This rigid statute creates a greater disadvantage to Indian AI developers as compared to many other parts of the world (e.g., Singapore and Japan) where TDM exceptions were enacted to support innovation.

Policy Paralysis: “Revenue Share” Proposal

The DPIIT (Department for Promotion of Industry & Internal Trade) created an expert committee during the summer of 2025 to review the Copyright Act and provide recommendations on how it should be amended due to the gap in legislation around AI and Copyright law. Recently, in December of 2025, reports indicate that the committee is likely leaning towards a “Revenue Sharing” model.

This proposed framework would allow AI developers to use copyrighted content when training their models, but would require them to pay a percentage of their revenue back to the copyright owners whenever they commercialise the AI product. While this seems like a fair way to handle it, many critics feel that this is simply “dead on arrival” for several reasons.

1. Attribution problem: It is technically difficult, if not impossible, to identify which specific training dataset caused which specific output from an AI model. How do you determine how much royalty should be paid to an individual contributor who wrote a three-sentence news summary from a model trained using 3 trillion tokens?
2. Check Post Raj: Within the tech industry, a government-mandated revenue-sharing scheme is seen as a return to the days of the “License Raj,” which could hinder the growth of domestic start-up AI companies that cannot afford the complex compliance processes that may be required under such a scheme.
3. Global disparity: If India places high royalty burdens on its AI companies while the EU and US do not, it could create an uneven playing field for Indian AI companies that will struggle due to their burden, while competitors will have a lower cost of entry into the market as compared to their Indian counterparts.

Moral Rights and Deepfakes

In addition to economic rights, it has been recently established that Moral Rights (Section 57) are also a concerning issue in 2025. In particular, the emergence of AI-created “deepfake” versions of popular figures (such as celebrities or politicians) has created an imminent need for legislation concerning personality rights.

For example, the Delhi High Court has issued “John Doe” orders (in the Anil Kapoor and Amitabh Bachchan cases) to protect popular actors from the unauthorised use of their image, but this was based on the right of publicity. The issue of copyright is still not clearly defined. If a machine were to produce a song using the same style as Arijit Singh, does this constitute a violation of copyright for prior works, or does this represent a form of imitation that does not qualify for copyright protection? Based on the current jurisprudential analysis within India, it would appear that “style” or “genre” is not deemed protectable under copyright law, which means that artists have no legislative recourse against generative imitators.

Conclusion: Way Forward

As we approach 2026, we can no longer take a “wait and see” approach. The case of ANI v. OpenAI will likely set the first judicial precedent; however, courts cannot create law.

For India to come up with an effective solution, there are 3 main areas to focus on:

1. Amending Existing Laws: We will need to put in place a specific exception for TDM for non-commercial research to assist universities and non-profit organisations.
2. Compulsory Licensing of Commercial Use of AI: Instead of having an ambiguous revenue share, we could set up a clear system, as is done with radio broadcasting, by which those who create copyrightable material can continue to receive compensation, while at the same time allowing for technology to progress rapidly.
3. Mandatory Disclosure of AI Datasets: As part of the development of an AI, AI developers must be legally required to disclose the datasets that were used to train the AI. Without this information, copyright holders will not even know if their rights are being violated.

India has aspirations to become a global hub for AI (“AI for All”), however, this aspiration should not come at the expense of India’s creative economy. The next few months for both the judiciary and legislative bodies will be focused on providing a way to walk this fine line; i.e., how does the machine learn, and how does the human copyright holder get compensated?

References

1. ANI Media Pvt. Ltd. v. OpenAI Inc., CS(COMM) 1028/2024 (Delhi High Court).
2. The Copyright Act, 1957, ss. 14, 52, and 57.
3. Kapoor, Surbhi, (2025) When Machines Create: The Emerging Copyright Challenges of Generative AI in India. Khurana & Khurana.
4. Pandya, Dyuti (2025) The Global South AI Copyright’s Test Case: India. Centre for European Policy Analysis (CEPA).
5. Economic Times, (2025, December 19). AI copyright, dead on arrival? Economic Times Opinion.
6. SpicyIP, (2025) The Tech is Dead, Long Live the Tech: Lessons in Indian SEP Litigation.
7. DPIIT Office Memorandum. (2025). Constitution of the Expert Committee on Generative AI and Copyright. Ministry of Commerce and Industry, Government of India.