Facts
ANI Media, an Indian news agency, sued OpenAI for copyright infringement. The suit and an application for an interim injunction were first listed on 19 November 2024. ANI made two claims: that OpenAI copied and stored its publicly accessible news to train the large language models (LLMs) behind ChatGPT, and that ChatGPT reproduced its works in its answers. It also asked for the stored works to be deleted. On the first day the court framed four issues and appointed two amici curiae. Publishers’ bodies and the music industry intervened for ANI, and an AI start-up and two policy groups for OpenAI. In October 2024 ANI had offered OpenAI a licence for its content for USD 7.5 million.
Question
The court addressed four issues. First, whether Indian courts had jurisdiction, and whether the Copyright Act 1957 could apply, given that training took place on servers in the United States. Second, whether ChatGPT’s answers amounted to memorisation and regurgitation, or substantial reproduction, of ANI’s news reports. The remaining two issues were taken together: whether storing the works for training infringed the reproduction right, and whether that use fell within the fair dealing exception in section 52(1)(a), in particular private or personal use, including research. A further question was whether a commercial use can qualify for that exception.
Decision
On 24 July 2026 Justice Amit Bansal dismissed the injunction application. Storing ANI’s works on US servers was the terminal step in a chain that began with access from India, so the court had jurisdiction on a prima facie view. The ChatGPT answers relied on by ANI post-dated training and did not show memorisation; they were generated by retrieval-augmented generation (RAG) and were not substantially similar to ANI’s reports. Section 52(1)(a) is not limited to non-commercial use, and research is not confined to humans but extends to machine learning. The use was limited to training, caused no proven market substitution and served the public interest. ANI could block crawlers, and its loss was compensable in money.
Why it matters
The judgment is described as the first substantive Indian ruling on training LLMs with copyright-protected news content. The court drew on US fair use decisions but based its analysis on a purpose test and a fairness test under section 52(1)(a). It stressed that its findings are prima facie and have no bearing on the final outcome of the suit. In weighing the public interest, it also took into account that developing an LLM would be economically unviable if training required licences from many sources, and that an injunction would harm LLMs being developed in India.
Related stages
ANI appealed to a Division Bench of the Delhi High Court. According to press reports, on 15 September 2026 the Bench issued notice to OpenAI, seeking its response, and listed the appeal for further hearing in December 2026.
The same question elsewhere
May AI be trained on copyright works?
Building the dataset fell within the text and data mining exceptions.
Kneschke v LAION e.V.United States · 2025In partTraining was fair use; building a library from pirated copies was not.
Bartz v. Anthropic PBCGermany · 2025NoMemorising lyrics in the model was reproduction, outside the exception.
GEMA v OpenAIUnited States · 2025NoUsing headnotes to build a competing product was not fair use (non-generative AI).
Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc.Storing news content to train the model was held prima facie to be fair dealing; an appeal is pending.
— this decision