Delhi High Court AI Training Ruling: A Win With Fragile Foundations

The Delhi High Court handed OpenAI a victory in the first Indian court ruling on AI training data, dismissing news agency ANI's request for an interim injunction. Justice Amit Bansal declared that training a large language model does not violate copyright.

The court adopted a human-learning analogy: because machines process text to extract information, they engage in research much like human scholars do. Since facts cannot be copyrighted, extracting information from a piece of content – separating it from the specific words used to express it – does not violate copyright, Justice Bansal wrote.

That reasoning would seem to settle the matter. If model training is no different from learning, no claim of copyright infringement should stand. The court, however, took a different path.

It concluded that ANI articles were stored temporarily during training, and since OpenAI did not obtain permission for that storage, the content was wrongfully copied. Yet the court said this fell within the fair dealing exemption under Section 52 of the Indian Copyright Act because all training is a form of research.

Rahul Matthan, a partner at Trilegal, laid out the implications in a column on the ruling. "Fair dealing is a defence, not a right," he wrote. "If we are going to justify AI training by claiming that even if it has been illegally copied, it is permitted under an exemption, we will have to establish that these grounds exist every time we embark on a training run."

Fair dealing under Indian law is tied to a narrow list of permitted purposes: research, criticism, and review. Unlike the open-ended "fair use" standard in the United States, an Indian AI company would have to continuously weigh the purpose of each use, the amount copied, and its effect on the market for the original work. "The operational uncertainty this will generate will be a precarious foundation on which to build an industry," Matthan said.

The border problem adds another layer. Fair dealing is a creation of Indian statute; its influence stops at the national border. The Regional Court of Munich held last month that AI music generator Suno violated copyright by training on a catalogue of German song lyrics. It handed a similar verdict against ChatGPT in November. "An AI industry that trains its models on the world's content and serves users everywhere cannot afford to ground the legality of its operations on the exemptions granted by one country only to have it denied by the next," Matthan wrote.

A deeper fragility, he argued, is that the court had to read the fair dealing provision "in a far more creative manner than its drafters contemplated" to reach its conclusion. While OpenAI succeeded in this case, "other courts in India may potentially take a less forgiving stance under different circumstances."

Matthan also challenged the court's finding that content is stored during training. He said the model retains only the statistical residue of what it has read, expressed as mathematical weights that have no bearing on the actual words of the article. "All that will be revealed is a collection of facts shorn of the form and expression that our copyright law protects," he said.

The ruling gives the Indian AI industry what Matthan called a "temporary reprieve from prosecution." What the industry needs, he said, is a clear and invariant rule – not a foundation built on the continued extension of an exemption.