The Delhi High Court on Friday held that OpenAI's use of original retrieved copyrighted works to train its large language models (LLMs) constitutes fair dealing under Indian copyright law and does not amount to copyright infringement.
The ruling came in a case filed by ANI Media Pvt Ltd in 2024 against OpenAI Inc and OpenAI OpCo LLC, alleging that the company infringed ANI's copyrighted material by using it to train ChatGPT.
The observation on fair dealing is the first of its kind by an Indian constitutional court.
Justice Amit Bansal said he was of the prima facie view that OpenAI's act of storing ANI's original retrieved works for training the LLMs underlying ChatGPT falls under Section 52(1)(a) of the Copyright Act and therefore does not constitute infringement.
The court also held that responses generated by ChatGPT using retrieval-augmented generation (RAG) do not infringe copyright because the outputs were not substantially similar to ANI's original content.
Justice Bansal said ANI had failed to establish that ChatGPT memorised or reproduced its copyrighted material. "ANI has failed to make a prima facie case for interim injunction," he said.
The court further ruled that the balance of convenience favoured OpenAI, observing that granting an interim injunction would cause irreparable injury not only to the company but also to the public at large.
ANI had argued that OpenAI unlawfully used its publicly available copyrighted content to train its models, reproduced its content verbatim in responses, and generated inaccurate or hallucinatory information falsely attributed to the news agency.
OpenAI countered that it did not store data directly, that copyright law does not prohibit using publicly available material for training in the manner alleged, and that its pre-training process took place outside India using data stored on foreign servers. It also said it had refined its models to prevent regurgitation of training data and had blocklisted ANI's domain from future training since November 2024.
Several media and publishing organisations are also parties to the case.