ANI Challenges Delhi HC Order Denying Interim Relief in Case Against OpenAI

The Delhi High Court's decision allowing OpenAI to teach ChatGPT on its journalism has been contested by ANI. On September 8, 2026, the appeal made it to a Division Bench presided over by Justice V. Kameshwar Rao, but the bench never convened. The new scheduled date is September 14.

The sole Indian decision on the subject of whether AI companies can train on copyrighted news is the order that is currently being appealed. At this stage, the decision is only preliminary. Additionally, it is the number that all Indian publishers are secretly using while negotiating licences. The value of the licence substitute is determined by it. In July of 2026, Justice Amit Bansal turned down ANI's request for an interim stay.

Why Earlier ANI’s Appeal got Rejected?

According to Justice Bansal, the training of the LLMs that underpin ChatGPT constitutes research that contributes to the advancement of artificial intelligence systems by producing new information. This places it under the purview of the Copyright Act, 1957, Section 52(1)(a), which grants an exception for fair dealing in relation to "private or personal use, including research". In addition, he clarified that the provision states that electronic storage of a work for such reasons does not constitute copyright infringement.

The court did not rule that OpenAI's commercial nature disqualified it automatically. The court ruled that ChatGPT's responses do not substantially replicate ANI's news stories. It was acknowledged that ChatGPT retrieves real-time data instead of reciting training data. A model's retention of its training text and its ability to emit it are both concerns of memorisation. Because of this, the court ruled that ANI has not proven that ChatGPT has caused any copying or paraphrasing of ANI's original literary works. The court maintained its authority over OpenAI, stating that it reproduces the product within its purview.

Content Copyright and Indian Laws

This ruling resolves a number of issues that a publisher would see as distinct. A news archive is acquired by an AI company. It keeps it safe. It uses it to change the weights of the models. It is uncertain whether the model will keep recoverable pieces. Whether or not the product emits these to the consumer is debatable. Indian copyright law takes a different tack on each of the five distinct issues. Storage is the subject of the Explanation to Section 52(1)(a).

Training, which involves making use of the stored copy rather than just storing it, is not addressed. The court overcame this by interpreting the identical clause to include training as research. The case's outcome depends on that reading. According to Section 52(1)(a)(i), "private or personal use, including research" is defined. No one's private or personal information is involved in a commercial model-training run. Now it's up to a division bench to decide whether it still fits inside that clause.

Based on what was available to him at the time, the sole judge provided an answer to the question posed at the prima facie stage. Commercially, as well as philosophically, the difference is significant. Each component—access, retention, and downstream output—is priced independently when a publisher licences an archive. A discovery that dismantles them eliminates the majority of the negotiable topics.