OPINION | India’s AI Copyright Challenge: Law, data, and doctrine limits

The frayed relationship between copyright and AI is seen globally and is not limited to India.

Generative AI systems rely on the blanket availability of data. They require training on large repositories of texts, images, audio, and videos, which regularly include works of publishers, journalists, artists, musicians, and other rightsholders. This fact has placed an unprecedented strain on copyright law. While existing doctrines, such as those relating to reproduction, the creation of derivative works, database rights, fair use (or fair dealing), and licensing contracts, could theoretically resolve many disputes, they were not drafted with industrial-scale automated training in mind. As such, practical gaps have been appearing around the world. Rightsholders are often unable to prove whether or how their work was used. Developers are left unclear over how to legally source their data. Training pipelines remain black boxes, and statutory carve-outs almost never contemplate large-scale commercial AI systems.

India is in a decisive position in this worldwide settlement, unlike other countries' jurisdictions. India has no specific text and data mining exception and no specific AI exemption under the Copyright Act, 1957. Thus, in such a case, mass-scale training with copyright materials without a licence could be a violation of Section 14, specifically the reproduction right, unless a delineation by the courts or legislature regarding what uses are legitimate is provided. This legal grey area, however, has become untenable since the generative models are moving from experimental to commercial applications.

AI and Copyright: The Challenge

The issue has already entered the Indian courtroom. The Delhi High Court is reviewing the case of ANI Media (P) Ltd. v. OpenAI Inc. (CS(COMM)-1028/2024), where the use of copyright-protected news content for large language model training has been questioned. The legal battle opens up the fundamental issues about infringement, the limits of fair use under Section 52, and the jurisdictional applicability in situations where training and deployment occur in different geographical locations. While the outcome remains pending, the litigation itself underscores a deeper problem: India is attempting to resolve twenty-first-century AI practices using a copyright statute built around human authorship and intentional copying.

Global Approaches to AI and Copyright

The frayed relationship between copyright and AI is seen globally and is not limited to India. In major jurisdictions, AI–copyright governance has largely evolved by way of adaptation instead of the creation of AI-specific copyright regimes. The U.S. courts and Copyright Office are guided by concepts like human authorship, fair use, and originality, thus indicating doctrinal continuity, although mixed litigation outcomes are common. The European Union too, takes its cues from the wording of the DSM Directive and its data mining exemptions, which are further augmented by the AI Act's transparency and opt-out provisions; nevertheless, these mechanisms were not originally intended for general-purpose AI and are still hotly debated. The UK's approach is being reconsidered in light of recent consultations, which indicates the government's unease with blanket training exemptions. Australia has clearly preferred the route of licensing and negotiated remuneration over wide TDM exceptions, while China's policy has been a pragmatic and state-directed one, managing the trade-off between flexible treatment of training-related copying and selective acceptance of AI-generated outputs.

India’s Struggle with AI Copyright and Liability

India's stance feels uneasy among the nations of the world in this regard. The Copyright Act, 1957 is silent on AI-generated content, artificial intelligence co-authorship, and large-scale automated analysis of protected works. The copyright system in India still roots itself firmly in human creativity and human liability. This situation raises doubts not only for the creators but also for the developers, who are left without proper guidance on permitted training methods.

The disparity becomes even more obvious when the application of AI in copyright matters is compared to other regimes of intellectual property. The Trademark Act, 1999 Section 27(2) keeps the passing-off action of common law as one of the rights against infringement, extending protection to unregistered marks or distinctive styles where confusion is caused by misrepresentation. On the other hand, the Designs Act, 2000 provides protection only to a new and original design, which is defined by its visual features and which can be applied through industrial processes. Hence, they are not clear-cut safeguards against the misuse of creative expression but rather assume identifiable human authorship and intent.

The Unresolved Issue of AI Liability

Matters are further complicated by judicial precedent. In Eastern Book Company v. D.B. Modak, the Supreme Court rejected the “sweat of the brow” doctrine and reaffirmed creativity as the threshold for copyright protection. But generative AI outputs are not the result of conscious skill or judgment; they are statistical products of training data and algorithmic computation. Similarly, the distinction between inspiration and imitation articulated in R.G. Anand v. Delux Films becomes difficult to apply when copying occurs without awareness or intent.

The most unresolved issue is liability. When an AI system infringes intellectual property, who is responsible? The developer, the deployer, the user, or no one at all? Indian copyright law presumes wilful human conduct. Yet generative systems can reproduce protected material without the knowledge or intention of any identifiable actor. This absence of accountability is not a minor defect; it is an emerging systemic risk.

A Call for Legal Reform

Courts have historically emphasised intent and likelihood of confusion, as seen in cases such as Coca-Cola Co. v. Bisleri International (P) Ltd., where unauthorised use that could easily mislead the public was considered infringement. It is getting more complicated to apply this reasoning to AI systems that have been trained on large amounts of data. The lack of legal protection for AI-produced works has the following two main effects: one is that such works are not covered by current law, and the other is that there is no clear mechanism for attributing responsibility when they violate existing rights.

The legal uncertainty has become a practical issue that can no longer be ignored. The generators of AI output, such as ChatGPT and Midjourney, are capable of creating content that can be very similar to already existing copyrighted works. However, Indian law lacks clarity on whether the user or the AI provider will be liable. What India requires is not the renewal of its copyright law but a closer alignment of copyright with its broader AI governance strategy. For there is one unmistakable trend worldwide: existing legal frameworks can work only when underpinned by transparency, institutional capacity, and clear policy signals.

If India gets this balance right, it will not only resolve domestic uncertainty but also contribute meaningfully to global conversations on responsible, innovation-friendly AI governance.

(Vidhi Sharma, Visiting Fellow at Future Shift Labs and Aastha Naresh Kohli - Advisor at Future Shift Labs & Lawyer at the Himachal Pradesh High Court.)

Views are personal, and do not represent the stand of this publication.