ANI has appealed the Delhi High Court order that allowed OpenAI to train ChatGPT on its journalism. The appeal reached a Division Bench headed by Justice V. Kameshwar Rao on September 8, 2026, but the bench did not assemble. It now stands listed for September 14, LiveLaw reported.

Why it matters: the order under appeal is the only Indian ruling on whether AI companies may train on copyrighted news. It is an interim, prima facie ruling. It is also the number every Indian publisher is quietly working from in licensing talks. It sets the value of the alternative to a licence.

What the single judge actually held

Justice Amit Bansal refused ANI’s interim injunction in July 2026. The court’s reasoning rested on four points.

  • Training is research. Justice Bansal held that “the process of training of LLMs underlying ChatGPT can be considered as a form of research for generating new knowledge and advancing artificial intelligence systems.” That brings it within Section 52(1)(a) of the Copyright Act, 1957, which exempts fair dealing for “private or personal use, including research.”
  • Storage is covered. The explanation to that clause says storing a work electronically for those purposes “shall not constitute infringement of copyright.”
  • Commercial use does not defeat the exception. The court declined to treat OpenAI’s commercial character as an automatic disqualifier.
  • The outputs did not reproduce ANI’s work. The court found that “responses produced by ChatGPT are not a substantial reproduction of the news articles of ANI.” It accepted that ChatGPT retrieves live information rather than reciting training data.

Memorisation is the concern that a model retains its training text and can be made to emit it. On that, the court held that “ANI has failed to satisfy this Court that any memorization and regurgitation of ANI’s original literary works has happened through the responses generated by ChatGPT.”

The court kept jurisdiction over OpenAI, noting that the “output is reproduced within the jurisdiction of this Court.”

Four acts, one finding

The judgment answers a set of questions that a publisher would treat as separate.

An AI company acquires a news archive. It stores it. It processes it to adjust model weights. The model may or may not retain recoverable fragments. The product may or may not emit them to a user.

Those are five different things, and Indian copyright law addresses them differently. The Explanation to Section 52(1)(a) speaks specifically to storage. It does not speak to training, which is a use of the stored copy rather than the storing of it. The court bridged that by reading training as research under the same clause.

That reading is the hinge of the case. Section 52(1)(a)(i) reads “private or personal use, including research.” A commercial model-training run is neither private nor personal. Whether it nonetheless sits inside that clause is what a Division Bench will now be asked. The single judge answered the question at the prima facie stage, on the record then before him.

The distinction matters commercially, not just doctrinally. A publisher licensing an archive is pricing access, retention and downstream output separately. A finding that collapses them removes most of what there is to negotiate over.

What ANI is asking for

ANI’s grounds are not yet public. LiveLaw’s report of September 8 records only that the appeal challenges the refusal of interim relief. MediaNama has sought a copy of the memorandum of appeal and will report the grounds once we have it.

The first is whether ANI attacks the research characterisation itself, or only its application to a commercial deployment. The second is whether the memorisation finding is challenged on evidence. The single judge held that ANI had not shown memorisation. He did not hold that memorisation does not occur. That is a difference a better evidentiary record could close.

ANI sued OpenAI in November 2024. The underlying suit remains pending, and the appeal concerns only interim relief.

Questions we have sent

MediaNama has written to ANI, OpenAI and the Digital News Publishers Association. We will update this post with their responses.

We asked OpenAI:

  • Does OpenAI treat the July finding as settling the position for Indian content?
  • Has anything changed in how companies acquire or use Indian publishers’ content since the order?
  • Does OpenAI hold licensing agreements with any Indian news publisher?

We asked ANI:

  • Which findings is the appeal directed at?
  • Does it challenge the characterisation of model training as research?
  • Has any AI company approached ANI for a licence since July?

We asked the Digital News Publishers Association:

  • Has the July order changed the terms AI companies are offering Indian publishers?
  • Does the DNPA propose any collective position on AI training rights?

Also read: