Use Of Copyrighted Material For Training Large Language Models: Indian Law After ANI V. OpenAI
Vatsal Sutaria, NMIMS' Kirit P Mehta School of Law, Mumbai
Archie Kotwani, NMIMS' Kirit P Mehta School of Law, Mumbai
ABSTRACT
Large language models are trained on very large collections of text, and much of that text is protected by copyright. This paper asks whether Indian law allows that use without a licence. It explains which steps in training involve copying, reads Sections 14 and 52 of the Copyright Act, 1957 against those steps, and examines the Delhi High Court's interim order of July 24, 2026 in ANI Media v. OpenAI. It then compares other jurisdictions and assesses the licensing model proposed by the Indian government in December 2025. The paper argues that the ANI order leaves two questions open, the source of training data and memorised output, and that reform in India should address those questions directly.
I. Introduction
A large language model (LLM) learns to predict the next word in a sequence by processing billions of words of text. Much of that text is under copyright, and its owners were rarely asked. Since 2023, owners have sued developers over this practice in the United States, the United Kingdom, Germany and India.
NASSCOM and Boston Consulting Group estimated in 2024 that the Indian AI market would reach about USD 17 billion by 2027. The legal question became concrete on July 24, 2026, when Justice Amit Bansal of the Delhi High Court refused an interim injunction sought by the news agency ANI against OpenAI. He held, prima facie, that storing ANI's articles to train ChatGPT fell within the fair dealing exception in Section 52(1)(a) of the Copyright Act, 1957.
