Co- authored by Shwetank Tripathi - Partner and Devangini Rai (Associate)

In a significant development for India’s emerging AI and copyright jurisprudence, the Delhi High Court, in ANI Media Private Limited v. Open AI OPCO LLC, CS(COMM) 1028/2024,(“Judgment”) delivered its judgment on 24 July 2026. The Court declined to grant interim injunctive relief to ANI Media Pvt. Ltd. (“ANI”) against OpenAI’s use of its copyrighted content for training large language models and generating AI-driven responses. The decision is among the first in India to directly examine how existing copyright principles apply to the training and operation of generative AI systems.

Background

ANI claimed that several of its copyrighted works being news articles were stored as part of OpenAI’s training data for LLMs which are infringing copies. Further, this data is reproduced by the LLMs in the form of responses to user queries. ANI produced specific instances of reproduction of its works by OpenAI wherein the original elements of ANI’s works were alleged to be memorised and retained in OpenAI’s LLMs which are then reflected in their responses. However as per OpenAI, it was undisputed that the training of OpenAI’s LLM Models ended in April 2022 - April 2024 which was prior to the publication of ANI’s cited instances of works being reproduced. 

Issues

Copyright infringement vis-à-vis storage / reproduction of ANI’s works

ANI contended that OpenAI converts its copyrighted works into vectors in the form oftraining data. Upon a user query, it predicts the response by matching the query with vectorsin its database, and re-converts the vectors into readable format, thus involving reproduction of ANI’s works. ANI further contended that the training data cannot be removed from LLMs’ retention once used and continuously gets reflected in the responses generated. In response, OpenAI argued that LLMs are only designed to learn information / facts from the training data to respond to user queries and not reproduce extracts of training data. Further, ANI had been unable to prove substantial reproduction of its works by OpenAI’s LLMs despite trying to artificially elicit verbatim reproduction of its works in responses to adversarial prompts. Regarding the allegation of storage of copyrighted date, OpenAI contended that storage of training data is an immediate step and the LLMs cannot access the training data once the training is complete.  

The Court held that that the responses generated by the LLMs do not involve memorisation of ANI’s works since it was undisputed that the LLMs producing such responses had been trained on data before the publication of ANI’s works in question. On the aspect of reproduction, the Court further held that in the case of news, copyright would only subsist in the form and manner of expression and not in the underlying facts. Hence, OpenAI’sresponses do not amount to a substantial reproduction of the expression occurring in ANI’s works as the responses rendered by OpenAI are summaries of ANI’s works without any exact copying. On storage of training data, the Court observed that as per the provisions of Copyright Act, 1957 (Act), specifically Section 14, temporary storage of training data by OpenAI would amount to an act of infringement unless it falls in any of the exceptions provided under Section 52 of the Act under which OpenAI was ultimately exempted, as discussed below. 

Fair Dealing Defence

ANI contended that usage of its copyrighted works does not fall under Section 52(1)(a)(i) of the Act as it is used by OpenAI purely for commercial purposes. In response, OpenAIcontended that storage of ANI’s works as part of OpenAI’s training data is protected under Section 52(1)(a) of the Act as it is being used by OpenAI privately for the purpose of research. Further, the incidental storage of infringing copies of literary works for the purposes as under Section 52(1)(a) is protected under the Explanation to Section 52(1)(a) and the caveat therein is only restricted to non-infringing copies of a computer programme. Further, use of ANI’s copyrighted works for training of LLMs is transformative in nature.

The Court held that Section 52 of the Copyright Act specifically provides for protection of non-commercial uses in some provisions. However, this distinction is visibly absent in Section 52(1)(a) of the Act. Thus, merely because OpenAI is a commercial venture, it does not prevent it from seeking exemption under Section 52(1)(a) of the Act. On the aspect of literal interpretation of the Explanation to Section 52(1)(a), the Court held that the caveat of incidental storage of a non-infringing copy is limited to a computer programme. The Court further held that since ANI’s copyrighted works are being used internally purely for the purpose of training, without the same being made available to the public, such use is privateand would be covered under ‘research’, within the meaning of Section 52(1)(a)(i). Further, the Court also held that while ANI primarily functions as a news agency, OpenAI offers services for content creation, research, translation and education etc. The Court observed that OpenAI uses ANI’s works purely for the purpose of training its LLMs. Pursuant to such training, OpenAI’s LLMs provide snippets of news articles, the content of which is different from the ANI’s copyrighted works being used for training the LLMs. Thus, the usage of ANI’s copyrighted works by OpenAI is fundamentally different from ANI’s business, whichdoes not affect ANI’s market for such works. 

Territorial Jurisdiction

OpenAI contended that since the training of data occurs on servers located in the USA, ANI is unable to establish the test of ‘purposeful availment’ with respect to the cause of action occurring outside India. Further, users using OpenAI’s service agree to its Terms of Use as per which all disputes are subject to the jurisdiction of the Courts of California. Further, Section 62(2) read with Section 1(2) of the Copyright Act was never meant to apply to causes of action occurring outside of India. In response, ANI argued that the physical location of OpenAI’s servers in the USA does not restrict ANI from bringing a suit against OpenAI since OpenAI offers its interactive and commercial services in India. Further, ANI’s principal and registered office is in Delhi which grants the Court territorial jurisdiction. 

The Court held that since ANI admittedly has its principal place of business and its registered office in Delhi, the Court has territorial jurisdiction as per Section 62(2) of the Act. Further, the Court also held that the training of data on foreign servers is a terminal step in the chain of events resulting in generating responses, which involves accessing copyrighted works created in India. Merely because the training of data occurs overseas, the same does not exempt the application of the Act with regards to the event as a whole. 

Comments

The Judgment marks an important early development in India’s evolving jurisprudence on the interaction between copyright law and artificial intelligence. Significantly, the Court recognised that the commercial use of copyrighted material may, depending on the facts, still fall within the scope of the “fair dealing” exception under Section 52(1)(a) of the Copyright Act. At the same time, the ruling should be read with some caution. Although the Court permitted intervention by stakeholders from other industries, its findings are confined to the facts before it and are based on a prima facie assessment of the evidence placed on record by ANI. The decision also arises from an interim application and remains subject to the outcome of the trial. Its broader significance will therefore depend on how these principles are applied in future disputes involving other categories of copyrighted works, including music, artistic works and other forms of creative content.