ANI’s OpenAI appeal slips to 14 September after the Delhi bench does not sit

Asian News International returned to the Delhi High Court on 8 September with an appeal against the July order that refused to stop OpenAI from using ANI copy to train ChatGPT. The division bench listed under Justice V. Kameswar Rao did not sit. Rao has been named Chief Justice of the Patna High Court. The appeal is now down for 14 September.

Justice Amit Bansal’s 24 July ruling is the first Indian judgment on whether an AI company may train on copyrighted news without a licence. He kept ANI’s copyright in the stories even though they sit free on the agency’s website. He also held that the Delhi court has jurisdiction because ChatGPT is offered to users in India. He then denied the injunction.

The decisive paragraph treats model training as research. Section 52(1)(a) of the Copyright Act, 1957, allows fair dealing for private or personal use, including research. An explanation to that clause says electronic storage for those purposes is not infringement. Bansal wrote that training the models under ChatGPT, by screening and organising stored literary works, can be treated as research that generates new knowledge. He also found that ANI had not shown ChatGPT outputs that reproduced its reports.

Two further points undercut urgency. ANI, he said, had placed no evidence that it had lost subscribers or syndication revenue. On 2 October 2024 the agency itself offered OpenAI a licence at US$7.5 million. The judge treated that figure as proof that any later win could be paid in money, which is the classic reason to refuse an interim ban. He added that an injunction would injure OpenAI and the public at large.

ANI’s appeal, filed through Sidhant Kumar of Unum Law, attacks the research label. The agency’s case is that publicly posted news is still a commercial product and that a company selling ChatGPT subscriptions is not doing private research. Publishers watching the file care about that sentence more than about ANI’s particular headlines. If training is research, the price of a licence is optional. If it is not, every Indian newsroom has a bill to send.

OpenAI’s defence at first instance was that facts are not protected, that learning from public text is transformative, and that a publisher can ask to be kept off the crawl list. The single judge did not need the whole of that argument once Section 52 did the work. The division bench will have to decide whether research in a 1957 statute covers the pre-training run of a commercial model.

The suit began in November 2024 and was the first Indian news copyright action against OpenAI. Parallel cases continue in the United States and Canada. Those courts are not bound by Bansal. Indian publishers are. Licensing talks now start from a published finding that storage for training is, at first look, lawful. That is why the 14 September date is not a scheduling footnote. It is the first chance to move that finding.

Publishers that have not sued are still using the July text as a price signal. If the division bench leaves the research finding in place, the alternative to a licence is free training plus a damages trial years later. If the bench narrows Section 52 to human scholarship, the next letters will go out with a number attached. ANI’s US$7.5 million offer, which hurt its injunction case, will then be read as a floor rather than as proof that money is enough.

Until then the operative law for anyone writing a news-to-model contract in India is the July order. Copyright remains with the agency. Jurisdiction sits in Delhi. An interim shutdown is off the table. Training looks like research. A US$7.5 million offer is on the record as a measure of harm. The appeal will have to break at least one of those holdings to change the market.