The Delhi High Court has refused to block ChatGPT in India, holding that OpenAI's storage of news agency ANI's copyrighted material for model training falls within the fair dealing exception of India's Copyright Act.
The holding
Justice Amit Bansal found that storage for LLM training falls under Section 52(1)(a), and that no prima facie infringement under Section 51 had been made out. The court also held that retrieval-augmented outputs were not substantially similar to ANI's reports, and that neither memorization nor regurgitation had been proven. An injunction, it said, would cause irreparable injury to the public at large.
Read the procedural posture carefully
This is an interim decision on ANI's injunction application. The main suit — CS(COMM) 1028/2024, filed in November 2024 — remains pending. Headlines saying the court found no copyright breach, or that it dismissed ANI's case, are describing something that has not happened. Order was reserved on March 27 and pronounced on July 24.
What ANI won
Almost every account omits it: the court ruled in ANI's favour on territorial jurisdiction, rejecting OpenAI's argument that an Indian court could not hear the claim.
What happens next
The suit proceeds to a trial on the merits, where the fair dealing analysis can be revisited against evidence rather than pleadings, and ANI can appeal the interim refusal. Nothing here is settled — it is an early signal from a court that will be hearing this case for some time.
Divergence
This is the first fair dealing holding on AI training by an Indian constitutional court, in the country with the largest ChatGPT user base — and it points the opposite way from the United States, where Anthropic recently settled an authors' training-data claim for $1.5 billion.