The recent judgment dated 24.07.2026 of the Hon’ble Delhi High Court in ANI Media Pvt. Ltd. Vs. Open AI OpCo LLC in CS(COMM) 1028/2024 dealt with innovations, such as, Large Language Models (LLMs) affecting creative works, such as, literary, dramatic, musical and artistic works, cinematograph films and sound recordings protected under the Copyright Act, 1957.
ANI, an Indian news agency, filed a suit against OpenAI alleging unauthorized use of its copyrighted news content by ChatGPT. ANI's case rested on two distinct claims:
ANI sought an interim injunction under Order XXXIX Rules 1 and 2 of the Code of Civil Procedure, 1908, restraining OpenAI from using its data and seeking deletion of ANI's content from OpenAI's training datasets.
The Court framed four issues; it addressed jurisdiction first and then the substantive copyright questions.
OpenAI argued that since its training servers are located in the United States, Indian courts lacked jurisdiction and the Copyright Act, 1957 could not apply extra-territorially. The Court rejected this, holding that although the models are trained overseas, ChatGPT's outputs are received and consumed within India. Hence, the Court held that the training and output contentions cannot be “entirely divorced” from one another.
Relying on the jurisprudential principles expressed in Neetu Singh v. Telegram[1],―in this case, the servers were located outside India, but content was circulated and infringement occurred within Indian jurisdiction―the Court held that in an age of cloud computing, conventional notions of territoriality cannot be rigidly applied.
On a prima facie basis, the Court held it had jurisdiction to entertain the suit.
ANI argued that ChatGPT's responses, generated through detokenization of vectorized training data, reproduced or closely mirrored its original articles, amounting to memorization and regurgitation of copyrighted text. It relied on illustrative examples from the plaint and on foreign authority, including:
The Court found the precedents distinguishable.
SUBSTANTIAL SIMILARITY: In general, Courts in other countries had held that non-transformative use of content does not constitute fair use. Contrarily, substantive summaries could not constitute copyright infringement. Further, the generation of hallucinated articles falsely attributed to publishers constituted a classic passing-off claim.
In the instant case though, substantial reproduction of ANI’s works had not been established. Critically, the Court noted that ANI's illustrations of allegedly infringing outputs were generated after OpenAI had already stopped training on ANI's data (following the dispute), undermining any inference that the outputs reflected memorized training content.
The Court also observed that news, unlike creative works such as song lyrics, involves reporting of facts rather than original creative expression. Consequently, the threshold for finding “substantial similarity” in expression is correspondingly higher.
In short, ANI failed to establish that ChatGPT's outputs constituted substantial reproduction of its works.
Hence, the output/reproduction claim failed at the interim stage.
ANI’s submissions hinged on the contention that scraping, storing, tokenizing, and vectorizing its articles for LLM training amounted to reproduction and/or adaptation of copyrighted works, regardless of purpose, and that public availability of its content online did not license unrestricted commercial use. It was argued that storage of any substantial part of a copyrighted work is infringement irrespective of purpose, and that there is no statutory text-and-data-mining exception under Indian law.
OpenAI's defence centered on the exceptions carved into copyright law, which exempt “fair dealing” with a literary work for purposes including private/personal use, research, criticism, or review. The Court undertook an extensive doctrinal analysis of fair dealing, its legislative history, and comparative jurisprudence, then applied a two-part purpose test and fairness test.
PURPOSE TEST: The Court held that commercial use of copyrighted material is not automatically disqualified ‘fair dealing’ protection. Further, the fact that large language models (LLMs) process training data and abstract therefrom, rather than retaining data as retrievable copies was duly noted.
The Court relied on the purpose of OpenAI's training activity; it concluded that extracting patterns, statistical relationships, and linguistic structures to train LLMs was distinct from storing or reproducing ANI's expression for redistribution. Such processing, therefore, could constitute “private or personal use” within the meaning of the applicable provision.
FAIRNESS TEST: The Court also applied a multi-factor fairness inquiry. It held that OpenAI's use of ANI's works was limited to training the LLMs underlying ChatGPT, not their wholesale re-publication. OpenAI’s use neither resulted in unfair economic competition with ANI nor caused actual/potential damage to ANI's business, since ChatGPT was not shown to be substituting for or substantially diverting ANI's own market.
The functions performed by ChatGPT— improving access to information, supporting education, scientific research, software development, translation, and accessibility tools for persons with disabilities — served a broader public interest, resonating with the constitutional and statutory commitment to advancing knowledge and learning.
Having found both the purpose test and the fairness test satisfied, the Court concluded, on a prima facie basis, that OpenAI's storage of ANI's works for LLM training constituted ‘fair dealing’ and therefore did not amount to infringement.
Even independent of the merits’ findings, the Court held the balance of convenience favoured OpenAI. It noted that ANI's claim was quantifiable in monetary terms, pointing to ANI's own prior offer to license its content to OpenAI for USD 7.5 million. This meant ANI could be adequately compensated by damages if it ultimately succeeded at trial.
By contrast, the Court determined that an interim injunction against OpenAI could cause harm not easily reducible to money, given ChatGPT's scale of use.
The Court placed significant weight on public interest, describing generative AI as transformative technology benefiting education, healthcare, financial services, agriculture, and skill development, and noted that India itself is developing indigenous LLMs. It held that requiring LLM developers to license data from every source individually would be economically unviable and stifle AI development, and that an injunction would adversely affect users, many of them non-paying.
On this basis, the Court declined to grant an interim injunction.
The Court held, on a prima facie basis, that:
The Court dismissed ANI’s interlocutory application seeking interim injunction.
However, the Court expressly clarified that these findings are prima facie observations for the purpose of the interim application only, and will not bind the final adjudication of the suit at trial.
At BLAZE VENTURES, we have elaborate processes and qualified professionals for advising inventors and enterprises on effectively creating and comprehensively protecting intellectual property assets.
_______________
[1] 2022 SCC OnLine Del 2637
[2] 931 F. Supp. 2d 537(SDNY 2013)
[3] Case No.42 O 14139/24,decided on 11.11.2025
[4] [2010] EWHC 3099 (Ch)
