Back
rules AI training on news content is , rejects ANI plea
In India's first substantive ruling on whether training large language models on copyrighted material amounts to infringement, the has dismissed news agency 's plea for an against . Justice Amit Bansal, in a detailed judgment delivered on , held that OpenAI's storage and use of ANI's news articles for training its ChatGPT models qualifies as "" for private or personal use including research under .
The Dispute: Two Claims, One Battle
ANI, one of India's largest news agencies, launched two distinct allegations against OpenAI. The first, a "training claim," alleged that OpenAI scraped, stored, and reproduced ANI's copyrighted works to train its LLMs without authorization. The second, an "output claim," contended that ChatGPT reproduced and communicated ANI's works to the public through responses generated for users. ANI relied on
, which confers the
to reproduce a work
"in any material form including the storing of it in any medium by electronic means."
Several intervenors joined the fray—the , , and supported ANI, while , , and backed OpenAI. Two , Professor Arul George Scaria and Mr. Adarsh Ramanujan, assisted the Court on the technical and legal complexities.
Jurisdiction: A Chain, Not a Single Link
OpenAI argued that since its servers and training operations were in the United States, the Copyright Act had no
. The Court disagreed at the
stage, reasoning that the acts of scraping ANI's data from India, transmitting it abroad, and generating outputs accessible within India formed one continuous chain of events. As the judgment observed, accepting OpenAI's argument
"would lead to evasion of Indian copyright law by infringers who shift the terminal link to servers abroad."
The Court also noted ANI's principal place of business lies within Delhi's jurisdiction, satisfying
.
When Retrieval Isn't Memory
ANI's output claim rested heavily on the theory of ""—that ChatGPT regurgitates training data verbatim. But here, the Court found a decisive factual flaw: the illustrative articles cited in the plaint were published in , well after the training cut-off dates of (GPT-4) and (GPT-4o). The responses ANI obtained, the Court determined, were more likely the product of —a technique where the model pulls live information from external sources rather than recalling stored training data.
Even on the merits of comparison, the Court found no . Applying the test from , the Court examined the works as a whole rather than dissecting selected portions. In the much-discussed Neeraj Chopra interview example, the Court observed that ANI had used adversarial prompts—including the word "exactly"—to elicit a specific response, and even then, ChatGPT's output added its own commentary and context rather than reproducing ANI's article.
The Court was also mindful that quotes from public figures like Neeraj Chopra's mother may not belong to ANI at all. Under , the person who delivers an address or speech in public is the of the copyright, and ANI produced nothing to show of such rights in its favour.
: A Purposeful Reading
On the training claim, the Court engaged in a two-step analysis under Section 52(1)(a): whether OpenAI's use served a permitted purpose, and whether it was "fair."
The Purpose Test. The Court rejected ANI's argument that commercial use disqualifies OpenAI from the defence. Pointing to the legislative history, Justice Bansal noted that Parliament deliberately excluded any non-commercial requirement from Section 52(1)(a) while expressly imposing it in provisions like Sections 52(1)(ad), (k), and (n). The Court also refused to confine "private use" to individuals, drawing on dictionary meanings and the 's decision in , which extended Section 52 protection to an educational institution.
Applying the "
," the Court reasoned that research
"is no longer confined to humans—it is now being done through Artificial Intelligence. However, ultimately the research is at the behest of humans and for the benefit of humans."
The training process, which involves machine learning through screening and organising stored works, was held to constitute a form of research.
The Fairness Test. Rather than importing the American four-factor test—which the Court noted has no statutory basis in India—Justice Bansal formulated a context-specific inquiry aligned with : whether the use conflicts with of the work, and whether it the author's .
Three factors emerged. First, OpenAI's use was limited to training its models, with no evidence of widespread
or regurgitation. Second, the functions of ChatGPT—content creation, translation, research assistance—are fundamentally different from ANI's news syndication business; the Court found no evidence of
or loss of subscription revenue. Third, the
weighs heavily:
"Training LLMs underlying ChatGPT therefore contributes to advancing scientific knowledge, developing innovative computational tools, disseminating information, promoting education, enhancing accessibility, and advancing artificial intelligence research."
Key Observations
"the illustrations given in the plaint are post the training of Open AI's LLMs and a case for of ANI's works on the basis of the said illustrations cannot be made out"
"Open AI's acts of storage of the literary works of ANI for the training of its LLMs would fall under Section 52(1)(a) of the Copyright Act and hence, would not amount to infringement"
" would be caused not only to Open AI but also to the public at large, if an is granted in favour of ANI at this stage."
"If an is granted, it would be detrimental to the growth of AI and more particularly, to the LLMs being developed in India."
What the Ruling Means—and What It Leaves Open
The injunction was dismissed on three grounds: ANI failed to establish a case, the favoured OpenAI, and irreparable harm would result to both OpenAI and the public. The Court also noted that ANI's own offer of a USD 7.5 million license to OpenAI demonstrates the claim is quantifiable and compensable in damages.
However, this is an interim order, not a final judgment. Several critical questions remain for trial. Whether "research" under Section 52(1)(a) was intended to cover industrial-scale machine processing, as opposed to human scholarship, is far from settled. The Court expressly declined to decide whether RAG-based outputs constitute "" under Section 2(ff), since ANI had not properly pleaded that claim. And the factual disputes around , market harm, and ownership of individual works remain live.
Notably, India lacks a statutory —a gap the European Union has addressed through opt-out mechanisms and the 's copyright compliance requirements. The 's generous reading of "research" bridges that gap by interpretation, but whether higher courts will sustain that interpretive bridge remains to be seen. For now, the ruling stands as the first substantive Indian precedent that AI developers, rightsholders, and litigants will have to reckon with.
