OpenAI scored a major AI copyright victory in India as a Delhi court said ChatGPT training on news agency ANI’s content did not break copyright law.
The Delhi High Court on Friday ruled that OpenAI’s use of ANI’s published material to train the large language models (LLMs) behind ChatGPT does not amount to copyright infringement at this stage.
The decision came after ANI, one of India’s largest news agencies, filed a lawsuit against OpenAI in November 2024, accusing the company of using its copyrighted reports without permission to train AI models. ANI also claimed ChatGPT had produced fabricated stories and incorrectly attributed them to the news agency.
Justice Amit Bansal rejected ANI’s request for an interim injunction that would have prevented OpenAI from using its content while the broader case continues. The court found that ANI had not shown evidence that ChatGPT had memorized or reproduced its news reports in responses generated for users.
Training data falls under research exception, court says
A key part of the ruling was the court’s interpretation of India’s Copyright Act. Justice Bansal said storing ANI’s articles for training the models behind ChatGPT falls under the law’s fair-dealing exemption for research.
“I am of the prima facie of the view that OpenAI’s act of storing ANI’s original retrieved works for training LLMs underlying ChatGPT falls under Section 52(1)(a) of the Copyright Act (fair dealing) and therefore does not amount to infringement…,” Justice Amit Bansal said in his ruling, according to The Indian Express.
The court also said ChatGPT outputs created through retrieval-augmented generation (RAG) were not substantially similar to ANI’s original content and therefore did not constitute copyright infringement. OpenAI argued that its models do not store training material as a directly searchable database of articles.
The company also said training data was stored outside India and that its models had been improved over time to reduce the chance of reproducing training material.
A landmark test for AI and copyright rules
The case marks the first major Indian court decision examining whether AI companies can train large language models using copyrighted news content without a license.
Similar disputes are ongoing in other countries, including lawsuits involving OpenAI in the United States and Canada. The outcome of these cases could influence how AI companies access online content and how publishers protect their work. The Delhi ruling, however, is not a final judgment on the entire dispute. ANI can continue pursuing its claims and may appeal the decision.
What the decision means for AI companies and publishers
For AI developers, the ruling provides some legal support for using publicly available copyrighted material for model training in India, at least while courts continue to define the boundaries of AI-related copyright use.
The decision could also make it easier for companies building AI systems in India to argue that requiring individual licenses for every piece of training data would make large-scale AI development difficult. The court noted that blocking AI training could affect the growth of language models and have broader public consequences.
For publishers and news organizations, the ruling raises concerns about how AI systems benefit from years of reporting, verification, and editorial work. News companies argue that AI-generated answers could reduce traffic to original reporting and weaken existing business models.
The road ahead for AI copyright rules
The case highlights a growing tension between encouraging AI innovation and protecting creators whose work helps power these systems.
While the Delhi High Court found no evidence that ChatGPT was simply reproducing ANI articles, future cases could focus more heavily on whether AI-generated content competes with the original creators or reduces the value of their work.
As AI companies continue training increasingly capable models, courts around the world will likely play a major role in deciding where the line falls between research, transformation, and unauthorized use.
Also read: Anthropic won final approval for a $1.5 billion AI copyright settlement after authors accused it of using unauthorized book copies to train Claude.