The judgment is India's first to examine AI companies' use of copyrighted works for foundation models, setting an analytical framework for Section 52 of the Copyright Act.
In a verdict giving interim relief to OpenAI, the Delhi high Court on Friday upheld using copyrighted content to train generative artificial-intelligence (AI) models, saying it amounted to 'fair dealing' under the Copyright Act and did not constitute copyright infringement.
A single-judge Bench of Justice Amit Bansal said that prima facie, the company's use of media agency ANI's copyrighted content to train its large-language models was permissible and within the scope of 'private or personal use, including research' under Section 52(1)(a)(i).
Key Points
- The Delhi high court upheld OpenAI's use of copyrighted content for training generative AI models, deeming it 'fair dealing' under the Copyright Act.
- Justice Amit Bansal dismissed ANI's plea for a temporary ban, stating the news agency failed to establish a prima facie case of copyright infringement.
- The court noted that ChatGPT's output, using Retrieval-Augmented Generation (RAG), was not substantially similar to ANI's original works.
- The ruling is expected to influence India's evolving debate on AI and copyright, though fundamental questions remain for future resolution.
Dismissing ANI's plea for a temporary ban on OpenAI's use of its material, Justice Bansal said the news agency had failed to establish a prima facie case of copyright infringement in relation to either OpenAI's training of its AI models or the responses generated by ChatGPT.
The larger question of whether OpenAI can continue using content makers' data without copyright infringement would continue to be heard as Justice Bansal sent the case back to the roster bench to decide which Bench would hear it.
The output generated by ChatGPT, using the Retrieval-Augmented Generation (RAG) technique, does not amount to infringement under Section 51 of the Copyright Act because it was not substantially similar to ANI's original literary works, he said in his judgment.
RAG is a technique that enhances AI responses by fetching facts from an external knowledge base before generating an answer.
"We welcome the court's decision. ChatGPT helps enhance human creativity and advance scientific discovery and medical research, and enables hundreds of millions of people in India and around the world to improve their daily lives," a spokesperson for OpenAI said.
"Our models empower innovation, and are trained on publicly available data and grounded in the principles of fair dealing."
Key Aspects of the Judgment
Justice Bansal also noted that ANI had failed to satisfy that memorisation and regurgitation of the agency's original literary works had happened through the responses generated by ChatGPT.
The judgment, of 135 pages, is the first in India to examine whether AI companies can use copyrighted works to train foundation models and lays down an analytical framework for applying Section 52 of the Copyright Act to AI systems.
The court also rejected ANI's contention that commercial entities were automatically excluded from claiming the defence, observing that Parliament had expressly limited certain copyright exceptions to non-commercial use but had not imposed such a restriction under Section 52(1)(a).
'The key to the success of an AI model is to access the information in the public domain. The development of LLMs and their success depend on availability of data. It would be economically unviable to develop an LLM if training an LLM would require licences from multiple sources,' the judgment said.
Jurisdiction and Future Implications
On jurisdiction, the court rejected OpenAI's argument that Indian courts could not examine the training claim because the company's servers were located in the United States. It held that accepting such a contention would enable copyright infringers to evade Indian law merely by locating servers abroad.
The ruling is likely to shape India's evolving debate on AI and copyright while leaving several fundamental questions to be settled either during trial or through legislation, experts said.
"While this is only an interim order, it has the potential to influence how AI developers and copyright owners approach the use of protected content for training AI models," said Ankit Sahni, partner at Ajay Sahni Associates and counsel for the Federation of Indian Publishers.
Ronil Goger, managing partner, Blaze Legal, said: "It is India's first substantive judicial engagement with AI and copyright, but it leaves fundamental questions unresolved, including whether AI training requires licences, how fair dealing applies to machine learning, and how creators should be compensated,"
For content companies, this underscores the need to adopt a broader strategy that does not rely exclusively on litigation.
