Delhi HC reserves order in ANI-OpenAI copyright dispute

The Delhi High Court has reserved its order on an interim application filed by Asian News International (ANI) seeking to restrain OpenAI from using its news content to train ChatGPT, in what has emerged as India’s first major judicial examination of the application of the Copyright Act, 1957 to artificial intelligence systems.

Justice Amit Bansal concluded hearings on the matter after a prolonged series of proceedings that began when ANI filed its suit on 19 November 2024. Since then, the case has seen 32 hearings, with arguments continuing until 27 March 2026. The Court’s forthcoming order on interim relief is expected to offer early judicial guidance on the legality of large-scale content ingestion for training generative AI models.

ANI has alleged that OpenAI used its copyrighted news articles without authorisation or licensing fees while training ChatGPT. According to the news agency, OpenAI deployed web crawlers to scrape content available on its website and incorporated the material into its training datasets. ANI further contended that responses generated by ChatGPT sometimes reproduced its articles, including content available only to subscribers.

The news agency argued that such use falls outside the “fair dealing” exemption under Section 52 of the Copyright Act, asserting that ChatGPT operates as a commercial product and that the use of journalistic works for AI training constitutes infringement. ANI maintained that infringement occurs at the ingestion stage itself, when copyrighted content is scraped, stored, and processed during model training.

OpenAI challenged the maintainability of the suit as well as the allegation of infringement. The company argued that neither it nor its servers are based in India, raising questions regarding the jurisdiction of Indian courts. It further submitted that the model is trained on publicly available material and learns statistical relationships between words rather than storing or reproducing expressive content.

The company also contended that ANI had not demonstrated instances of verbatim reproduction in ChatGPT outputs. OpenAI stated that the training process extracts what it characterised as “non-expressive” features of text, which it argued fall outside the scope of copyright protection. During the course of proceedings, OpenAI informed the Court that it had blocked ANI’s domain from future training.

The dispute expanded into a broader industry debate, with six intervenors participating in the proceedings. Organisations including the Digital News Publishers Association, Indian Music Industry, and the Federation of Indian Publishers supported ANI’s position, arguing that AI training processes such as scraping and tokenisation involve copying and transforming protected expressions.

These intervenors contended that intermediate copies created during the training process constitute infringement under Indian copyright law and warned that AI-generated outputs could act as substitutes for original works, causing economic harm to creators and publishers.

On the other side, intervenors including the Broadband India Forum, Flux Labs AI Private Limited, and IGAP Project LLP supported OpenAI’s position. They argued that large language models do not reproduce copyrighted works in material form and that summarising information from publicly accessible sources does not amount to copyright infringement.

The Court also examined the technical functioning of large language models, including processes such as tokenisation and temporary storage of data during training. While ANI argued that Indian copyright law does not permit non-infringing intermediate copies beyond limited exceptions for transient storage, OpenAI maintained that such storage is temporary and necessary to extract statistical patterns from text.

Court-appointed amici curiae presented divergent views on the issue. One expert suggested that OpenAI’s training processes could amount to copyright infringement, while another opined that temporary storage of works during machine learning training may fall within permissible limits.

The Court’s order on interim relief is expected to address key questions surrounding whether intermediate copies created during AI training qualify as transient storage under copyright law and whether existing statutory provisions adequately address machine learning technologies.

The decision could significantly influence how Indian courts balance copyright protection with technological innovation in the development of generative artificial intelligence.