Lawsuit tracker

High Court of Delhi, before Justice Amit Bansal · Active

ANI v. OpenAI

India’s biggest wire service is asking the Delhi High Court to rule that training a chatbot on copyrighted news is itself infringement — a question tangled up with whether foreign AI companies can be sued in India at all.

Key facts

  1. Filed in the High Court of Delhi in November 2024 by Asian News International (ANI Media) against OpenAI.
  2. ANI alleges infringement both at the training stage — copying, tokenizing and storing its wire copy — and at the output stage, including fabricated stories attributed to ANI.
  3. OpenAI blocklisted ANI’s domains from crawling in October 2024 but continues to contest the case on the merits and on jurisdiction.
  4. The interim injunction hearing ran 32 sessions over sixteen months and concluded March 27, 2026; Justice Amit Bansal reserved orders on April 2, 2026.
  5. India has no US-style fair use; the dispute turns on the closed list of exceptions in Section 52 of the Copyright Act, 1957.
PartiesANI Media Private Limited (Asian News International) v. OpenAI entities
CourtHigh Court of Delhi, India
DocketANI Media Pvt. Ltd. v. OpenAI (High Court of Delhi, before Justice Amit Bansal)
Filed2024-11
Content typejournalism
StatusAs of July 2026, orders reserved. After 32 hearings ending March 27, 2026, Justice Amit Bansal reserved his ruling on ANI’s interim injunction application — expected to be India’s first substantive word on AI training and copyright.

What ANI alleges

ANI Media, India’s largest news agency, argues that OpenAI’s infringement does not wait for an output. Under its theory, the moment ChatGPT’s training pipeline scrapes, tokenizes and stores ANI’s copyrighted wire articles, a copy has been made — whatever the model produces later. ANI adds a second, separate claim: that ChatGPT sometimes fabricates quotes and stories and attributes them to ANI, which the agency says damages its reputation for accuracy as much as it infringes its copyright. It seeks roughly ₹2 crore in damages and an injunction against further use of its content.

OpenAI’s defense

OpenAI’s answer rests on three legs. First, a technical one: large language models learn statistical relationships between tokens, not stored copies of expression, so training does not itself reproduce protected material. Second, a mitigation one: it blocklisted ani.in and related domains from its crawlers in October 2024, after the suit was filed. Third, a jurisdictional one: OpenAI has no servers or training infrastructure in India, so Delhi courts lack authority over conduct that happened abroad.

Why jurisdiction is its own battle

Before the court can reach fair dealing or infringement, it has to decide whether it can hear the case at all. In February 2025, an amicus curiae appointed by the court argued Delhi does have jurisdiction, reasoning that harm to an Indian rights holder from content reaching Indian users can be litigated where the harm lands, not only where the servers sit. That question matters well beyond this case: if it goes ANI’s way, any foreign model company whose outputs reach Indian users could face suit in Indian courts.

Where the case stands

The court framed four core questions: training-stage copying, output infringement, whether Section 52’s fair-dealing exceptions cover AI training, and jurisdiction. Arguments on ANI’s interim injunction application ran across 32 hearings between November 2024 and March 27, 2026 — an unusually long fight over interim relief — before Justice Bansal reserved orders on April 2, 2026. Industry groups including the Federation of Indian Publishers and the Digital News Publishers Association sought to intervene in support of ANI in early 2025.

Why it matters for training-data licensing

Section 52 is a closed list of exceptions, not an open-ended fairness test like the US fair-use factors — there is no obvious slot for “the model needed the text to learn language.” If the court finds AI training does not fit any listed exception, that reading would leave companies operating in one of the world’s largest English-language markets with no statutory defense at all, pushing them toward licensing news content rather than litigating case by case. It would also confirm that opting a publisher out after the fact, as OpenAI did with ANI’s domains, does not retroactively cure unauthorized use of what was already taken.

What to watch

  • Justice Bansal’s reserved ruling on ANI’s interim injunction, which could set the first Indian precedent on AI training and copyright.
  • Whether the court resolves jurisdiction as a threshold matter or folds it into the merits.
  • Whether the Federation of Indian Publishers and Digital News Publishers Association interventions broaden the case’s scope.
  • Any licensing arrangement that could narrow or moot the dispute before a final merits ruling.

Sources

copyrightjournalismIndiafair dealingjurisdictiontraining data

See something wrong? Send a correction.

Jaeden Schafer

Jaeden Schafer

Jaeden Schafer is the founder of fiund and host of the AI Chat podcast. He covers the training-data market and the lawsuits shaping it.

Want data that clears this in diligence?

Whether you're building a model or sitting on an archive, the first conversation is short and specific.

Send a brief