Who Owns Data? Delhi HC Reserves Order in Landmark ANI vs OpenAI Dispute

The Delhi High Court has concluded hearings on ANI's interim application seeking to prevent OpenAI from using its content to train ChatGPT. Justice Amit Bansal has reserved his order, marking India's first legal examination of applying the Copyright Act, 1957, to AI training. The case was filed on 19th November 2024, and till date, a total of 32 hearings have taken place between 19.11.2024 and 27.03.2026. ANI filed the case in 2024, alleging that OpenAI used copyrighted news content without permission or licensing fees. It objected to OpenAI's use of web crawlers, claiming that ChatGPT copied and used its content for training and reproduced its articles in response to user queries. ANI argued this usage is not covered by the “fair dealing” exemption,

as ChatGPT’s responses are commercially driven. OpenAI challenged both the court's jurisdiction, based on its claim that it, as well as its servers, were not based in India, and the claim of infringement, arguing that its model accesses publicly available content, such as ANI’s content available on its website, and uses what it claimed were ‘non-expressive’ data elements outside the scope of copyright. OpenAI also claimed that ANI failed to provide verbatim reproductions and that it had since blocked ANI’s domain from future training. The dispute became a sector-wide matter, with six intervenors participating. The Digital News Publishers Association, Indian Music Industry, and Federation of Indian Publishers argued in favour of law that protects copyright, while Broadband India Forum, Flux Labs AI Private Limited, and IGAP Project LLP sided with OpenAI. ANI contended that infringement occurs at the ingestion stage itself. ANI highlighted that some content found in ChatGPT's outputs was restricted to subscribers. It emphasised ownership over its journalistic works and argued that public availability does not grant rights for commercial use. ANI, DNPA and IMI asserted that LLM technical processes, such as scraping and tokenisation, involve storing and transforming protected expressions, potentially leading to infringement at multiple stages of model development. In contrast, OpenAI maintained that LLMs learn statistical relationships rather than expressive content, and that any storage is temporary and intermediate. ANI argued that Indian copyright law does not allow non-infringing intermediate copies beyond exceptions for transient storage. OpenAI countered that temporary storage used to extract what it categorised as ‘non-expressive features’ does not infringe copyright. The interpretation of the fair dealing exemption in Section 52 was hotly contested. OpenAI claimed LLM training is private research, not public use, while ANI resisted expanding exceptions for commercial AI systems. Intervenors like DNPA and IMI argued that digital storage exceptions are limited to non-commercial libraries, and commercial-scale copying is deliberate infringement. They also asserted that AI-generated outputs act as market substitutes, causing economic harm. Intervenors aligned with OpenAI advocated for a legal interpretation which would favour Generative AI, stating that LLMs don’t reproduce content in material form and that summarising public-domain data does not infringe copyright. Retrieval-Augmented Generation (RAG) functionalities were discussed as non-infringing. The Court will likely address substantive copyright issues, including whether intermediate copies during training qualify as transient storage and if current laws adequately cover machine learning systems. Court-appointed amici curiae provided divergent views: one advised OpenAI’s actions constituted infringement, while another argued storage for training is permissible. The Court’s decision on interim relief will provide early guidance on how India's legal system may approach large-scale content ingestion for generative AI, influencing the balance between copyright protection and machine-learning innovation in India.