ANI appeals Delhi High Court's AI training

ANI has appealed a Delhi High Court order that permitted OpenAI to train its ChatGPT model on the news agency's copyrighted journalism. The appeal was filed on September 8, 2026, before a Division Bench headed by Justice V. Kameshwar Rao.

The order under appeal is the first Indian judicial ruling on whether artificial intelligence companies can use copyrighted news material for model training. Justice Amit Bansal issued the interim, prima facie ruling in July 2026, refusing ANI's request for an injunction against OpenAI. This decision is currently shaping licensing negotiations between Indian publishers and AI firms.

The court's reasoning for refusal

Justice Bansal's refusal of interim relief was based on four key points. He held that the process of training large language models like ChatGPT could be considered a form of research aimed at generating new knowledge and advancing AI systems. This characterization brings the activity within the scope of Section 52(1)(a) of India's Copyright Act of 1957, which exempts fair dealing for private, personal use, or research.

The court also noted that an explanation within that clause states that storing a work electronically for such purposes does not constitute copyright infringement. It declined to treat OpenAI's commercial nature as an automatic disqualifier from using the research exception. Also, the judge found that ChatGPT's outputs did not substantially reproduce ANI's original news articles, accepting OpenAI's position that the model retrieves live information rather than reciting memorized training data.

On the specific issue of memorization, where a model might retain and emit training text, the court stated ANI had failed to provide sufficient evidence that this had occurred through ChatGPT's responses. The court retained jurisdiction over OpenAI, noting that the model's output is reproduced within India.

The commercial hinge of the case

The judgment effectively consolidated several distinct acts-acquisition, storage, processing for model weights, potential retention of fragments, and possible emission to users-under a single legal finding. Indian copyright law addresses these acts differently. The legal explanation regarding storage does not explicitly address the subsequent act of training, which is a use of the stored copy.

Justice Bansal bridged this gap by reading the training process itself as research under the same copyright clause. This interpretation is the central hinge of the dispute. Section 52(1)(a)(i) covers "private or personal use, including research." A commercial model-training operation is neither private nor personal. Whether it can still fall under the research exception is the core question the Division Bench will now consider.

This distinction carries significant commercial weight. When a publisher licenses an archive, it typically prices access, retention, and downstream output as separate elements. A legal finding that collapses these distinct stages removes much of the substance from licensing negotiations.

ANI's grounds and pending questions

The specific legal grounds for ANI's appeal are not yet public. LiveLaw's report indicates the appeal challenges the refusal of interim relief. MediaNama states it has requested a copy of the memorandum of appeal and will report the details once obtained.

Key questions remain about the appeal's focus. It may challenge the characterization of model training as research itself, or it may argue against applying that exception to a commercial deployment. Another potential line of attack concerns the evidence around memorization. The single judge ruled that ANI had not demonstrated memorization occurred; he did not rule that it cannot occur. A stronger evidentiary record could alter this finding.

ANI originally sued OpenAI in November 2024. The underlying lawsuit is still pending, with this appeal dealing solely with the interim injunction.

MediaNama's queries to stakeholders

MediaNama has sent questions to ANI, OpenAI, and the Digital News Publishers Association (DNPA), promising to update its report with their responses.

The outlet asked OpenAI if it considers the July ruling as settling the legal position for using Indian content, whether its practices for acquiring or using publisher content have changed since the order, and if it holds any licensing agreements with Indian news publishers.

Questions to ANI probe which specific findings the appeal targets, whether it challenges the "research" characterization of model training, and if any AI company has approached the agency for a license since July.

MediaNama asked the DNPA if the court order has altered the terms AI companies are offering to Indian publishers and whether the association is proposing a collective position on AI training rights.