E-Paper
Delhi HC's OpenAI-ANI ruling: What it means for AI, copyright and publishers in India — so far
The court ruled on three key questions — whether training is fair dealing, whether outputs infringe, and whether Indian courts can hear the case.
The Delhi high court has ruled that OpenAI did not break Indian copyright law by using news agency ANI's material to train ChatGPT — the first Indian judgment to test how the country's 1957 copyright statute applies to generative AI.
The decision, delivered on July 24 by Justice Amit Bansal, could potentially set the terms of engagement between AI firms, publishers and rights holders for years to come. Here is what the 135-page verdict says, and why it matters.
What was the case about?
ANI sued OpenAI for copyright infringement, arguing the company commercially exploited its reportage to train its large language model (LLM) without a licence. The suit seeks a permanent restraint on the use of its content, deletion of the stored material and ₹2 crore in damages.
The order last week dealt with ANI's interim application, which sought to stop OpenAI from further use of the agency's work pending the full trial. The broader suit will be heard starting September 11.
The proceedings drew interventions from the Indian Music Industry, the Federation of Indian Publishers and the Digital News Publishers Association — a measure of how much publishers, musicians and news organisations see riding on the outcome.
The court also appointed two experts to help it navigate the technology: Professor Arul George Scaria of the National Law School of India University, who took the view that storing ANI's data to train the model was not infringement, and advocate Adarsh Ramanujan, who argued that copying data without consent was infringing unless justified as fair use.
Can Indian courts even hear the case?
OpenAI's first line of defence was jurisdictional. Its LLM, the company argued, was trained outside India on servers stored abroad, which meant that California courts alone could decide the matter and the Indian Copyright Act did not apply.
ANI countered with Section 62(2) of the Act, which lets a copyright owner sue where it resides or carries on business.
Justice Bansal agreed. Storing the works on US servers, he held, was only the final step in a chain that began with copyrighted material being accessed in India. "On a prima facie view, Open AI's argument is untenable as it would lead to evasion of Indian copyright law by infringers who shift the terminal link to servers abroad," the court said.
Also read: AI and copyright: Need for a nuanced approach
Does ChatGPT's output infringe copyright?
ANI argued that OpenAI's model memorised its copyrighted content and that ChatGPT reproduced it in response to user prompts.
OpenAI countered that its models do not store third-party text but generate fresh responses designed to avoid regurgitation, and that ChatGPT functions merely as an interface between users and content.
The court sided with OpenAI on this point.
Outputs generated by ChatGPT using the retrieval-augmented generation (RAG) technique were not substantially similar to ANI's original literary works, it held, and therefore did not amount to infringement under Section 51 of the Copyright Act.
Is training AI model on copyrighted work 'fair dealing'?
This was the heart of the case. ANI argued that OpenAI's use was purely commercial and could not qualify for the "fair dealing" exception, which under Section 52(1)(a) shields private or personal use, research, criticism, review, the reporting of current events and current affairs, and the use of public lectures.
OpenAI's position was that storing material to train ChatGPT was private use by a private entity, regardless of the commercial nature of the eventual product.
The court accepted OpenAI's framing.
The training data, Justice Bansal reasoned, was accessible only to the model itself, and not to the public for viewing or downloading.
"Open AI stores the literary works in a closed space without access to the public. The data obtained by the LLMs for training purposes is used for private purposes. The said data is accessible only to the LLM models themselves. The said data is not publicly available to any human entity either for access or for download. Therefore, in my opinion, the use amounts to being purely private," he said.
Both the "purpose test" and the "fairness test" under the provision were satisfied, the court added, and the commercial character of OpenAI's operations did not by itself bar the defence.
Also read: The Aziz-Dube saga: A sleight of copyright?
Why did the court refuse an interim injunction?
Beyond the copyright question, Justice Bansal weighed the practical fallout of restraining OpenAI at this stage.
Granting an interim injunction, the court said, would hurt the growth of artificial intelligence in India and adversely affect millions of ChatGPT users, many of whom do not pay for the service.
That balance-of-convenience finding could be cited by AI firms defending a training-data suit in India from here on.
How does this fit with rulings abroad?
Courts in other jurisdictions have taken a harder line.
Last month, a US federal judge granted final approval to a $1.5 billion settlement in Bartz vs Anthropic — the largest copyright recovery in American history. An earlier order in the case found that while training AI on legally acquired books could count as fair use, building datasets from pirated "shadow libraries" could not; the settlement requires Anthropic to pay roughly $3,000 for each affected work and destroy the pirated material.
In 2025, a Delaware court in Thomson Reuters v. Ross Intelligence held that Ross's use of Westlaw content to train a competing legal-research AI was not fair use — the first US ruling to reject the fair-use defence in an AI case.
In Europe, France's competition authority fined Google €250 million in 2024 for breaching prior transparency commitments to publishers after training its Bard chatbot on their content without notice.
