OpenAI wins Delhi HC interim ruling in ANI copyright case
TL;DR
- The Delhi High Court on July 24 refused ANI's interim injunction, holding OpenAI's storage of ANI articles to train ChatGPT falls under Section 52(1)(a)(i).
- Justice Amit Bansal's 135-page judgment is the first substantive Indian court finding on training large language models with copyrighted news content.
- The court said ANI failed to show ChatGPT memorised or regurgitated its work; the main suit and hallucination-attribution claim remain pending.
An Indian court has given OpenAI its first substantive win on the question every AI lab and every publisher is fighting over in parallel jurisdictions, namely whether training a large model on a news outlet's back catalogue is fair use. According to reporting on the ruling, Justice Amit Bansal of the Delhi High Court on July 24, 2026 refused to grant news agency ANI an interim injunction against OpenAI, holding at the prima facie stage that OpenAI's storage of ANI's articles to train ChatGPT falls within the fair-dealing exception in Section 52(1)(a)(i) of India's Copyright Act.
This is the first substantive Indian court finding on AI training, and per Business Standard the 135-page judgment lays down an analytical framework for applying Section 52 to AI systems. Practically, it puts the burden on the publisher to prove that the model memorised or regurgitated protected work. On Justice Bansal's reading, ANI failed to satisfy the court that ChatGPT had reproduced its reports in generated responses.
ANI's original complaint, filed in November 2024, went further than the pure training-data argument. The agency also alleged that ChatGPT hallucinated stories and attributed them to ANI, which it argues can damage its reputation and lead to public disorder. That piece is not resolved here. The ruling addresses only the injunction request, and the main suit remains pending.
The honest caveat is that this is preliminary. The court itself said its observations are limited to the interim application and have no bearing on the final outcome, and Section 52 has not previously been tested against foundation-model training, so an appellate court could still land differently. What the reporting does not give you is how the bench will treat the hallucination-and-attribution claim on the merits, or whether the fair-dealing reading will hold once fuller evidence about how ANI's archive was ingested is on record.
For OpenAI, and for every other lab training on scraped news, this is a precedent worth citing in the parallel suits working through courts in the United States and Canada. For Indian publishers, it is a signal that a licensing deal, rather than an injunction, is the realistic route to being paid for their archives.
Shared on Bluesky by 2 AI experts
-
OpenAI wins first battle in ongoing Indian copyright lawsuit A fascinating ruling from the High Court of Delhi in India regarding the very high-profile case of ANI Media v OpenAI (decision here). Asian News Internationa…
View on Bluesky →
Originally reported by drive.google.com
Read the original article →Original headline: ANIMedia.pdf