OpenAI vs ANI: Court's interim relief to OpenAI raises bar for publisher copyright claims

The interim order refusing ANI an injunction against OpenAI raises the evidentiary bar for publishers, shifting the immediate battle towards demonstrable reproduction, market harm and commercial licensing.

The Delhi High Court, on Friday, delivered India’s clearest judicial signal yet on whether artificial intelligence companies can use copyrighted news to train their models, holding at the interim stage that OpenAI’s storage of ANI’s articles for training ChatGPT falls within the fair-dealing exception under Indian copyright law.

The court also found that the ChatGPT outputs placed before it were not substantially similar to ANI’s original articles and that the news agency had failed to establish any memorisation or regurgitation of its copyrighted material.

Read: Delhi HC denies ANI interim relief against OpenAI in ChatGPT copyright case

Consequently, Justice Amit Bansal dismissed ANI’s application seeking an interim injunction against OpenAI. However, the court expressly clarified that its observations were limited to the interim application and would have no bearing on the final outcome of the lawsuit. 

That qualification matters. The order does not conclusively declare all training on publicly available copyrighted material legal. But it changes the immediate terrain of India’s first major publisher-versus-generative-AI copyright contest.

For news publishers, merely showing that their material may have entered an AI company’s training pipeline may no longer be sufficient. The court’s analysis suggests they may also need to prove that protected expression was substantially reproduced, that the model memorised or regurgitated their work, or that its outputs caused identifiable harm to the market for the original content.

“The Delhi High Court’s order is only an interim, prima facie ruling, so it does not finally settle the legality of training AI models on copyrighted news content in India,” said Alay Razvi, Managing Partner at Accord Juris.

“At this stage, the court was not persuaded that ANI had made out a strong enough case for immediate injunction against OpenAI. The practical effect is that the dispute continues to trial, while the status quo remains in place.”

According to Razvi, the order indicates that allegations of AI-training infringement will require concrete evidence rather than broad claims about content use.

“For publishers, the order is important because it shows that allegations of AI training infringement will need concrete proof of copying or identifiable regurgitation, not just broad claims of use. But it is not a final precedent. The broader questions on fair dealing, infringement and AI training remain open for definitive judicial determination,” he said.

ANI had argued that OpenAI copied and stored its copyrighted news content without permission to train its large language models and then used that material to generate responses through ChatGPT. OpenAI maintained that facts cannot be monopolised and that publishers can block its crawlers when they do not want their websites accessed.

The court drew a distinction between ANI’s business of gathering and syndicating news and the broader functions performed by ChatGPT. It observed that the AI tool generates summaries or snippets in language different from ANI’s original expression and, in the material examined at this stage, did not operate as a substitute for ANI’s articles.

It also noted that ANI had not placed evidence on record showing that OpenAI’s activities had caused it to lose subscribers, market share or syndication revenue. On a prima facie assessment, the court therefore found no market substitution or actual or potential damage to ANI. 

Soumya Singh, Co-Founding Partner at Thistle&Law, described the order as more consequential than routine preservation of the status quo.

“This does not finally settle the ANI-OpenAI dispute for Indian publishers because it is still an interim order, but it is far from routine status quo preservation,” she said.

“The Court has given India’s first meaningful prima facie signal that training use may fall within fair dealing where there is no substantial reproduction, memorisation or regurgitation in outputs. For publishers, the battle now shifts from merely objecting to ingestion towards proving market harm, protected expression being reproduced, or unfair substitution at trial.”

The court’s reasoning on public interest may prove equally significant for India’s emerging AI industry. It said large language models depend on access to data and that requiring developers to obtain licences from numerous individual sources could make their development economically unviable.

An interim injunction, it added, could hinder the development of Indian LLMs and adversely affect millions of ChatGPT users in the country. The balance of convenience, in the court’s assessment, therefore weighed against ANI. 

For publishers, however, the commercial question remains unresolved even where the immediate copyright argument has weakened.

AI companies including OpenAI have entered licensing and partnership arrangements with selected news organisations internationally, creating a market in which some publishers are compensated for access while others remain outside the negotiating room. ANI itself had offered OpenAI a licence for its content for $7.5 million, according to the order. The court treated that offer partly as evidence that any eventual injury could be quantified and compensated through damages. 

Gopa Menon, co-founder and COO of theblurr, said the order could force publishers to look beyond litigation as their principal defence.

“The ruling doesn’t end the conversation on AI and publisher rights, but it does shift the focus. Publishers can no longer rely only on litigation; they will need to strengthen their value through licensing, proprietary content, and AI-friendly commercial partnerships,” he said.

“The leverage now comes from differentiation, not just ownership.”

That may be the less comfortable implication of the ruling for the media business. Copyright ownership remains intact, and the final trial could still produce a different outcome. But unless publishers can demonstrate reproduction or economic substitution, ownership alone may not compel an AI platform to pay.

The next phase of the dispute is therefore likely to unfold across two parallel tracks: inside the courtroom, where ANI must substantiate infringement and market harm, and outside it, where publishers must decide whether their strongest negotiating asset is the right to exclude AI systems or the commercial value of being included in them.