On 24 July 2026, the Delhi High Court delivered a significant interim order in ANI Media Pvt. Ltd. v. OpenAI Inc., marking one of the first substantive judicial pronouncements in India examining the intersection of copyright law and artificial intelligence (AI) model training. While the underlying suit remains pending, the Court's observations provide important guidance on how Indian courts may approach copyright claims arising from the development and deployment of large language models (LLMs).
Background
The dispute arose from a suit filed by ANI Media Pvt. Ltd., one of India's leading news agencies, alleging that OpenAI had trained ChatGPT on ANI's copyrighted news content without authorization. ANI further claimed that ChatGPT could reproduce portions of its reporting and, in certain instances, generate fabricated statements that were incorrectly attributed to ANI, thereby causing reputational harm.
OpenAI challenged the maintainability of the proceedings before the Delhi High Court, arguing that the training of its AI models did not occur in India and that its servers are located outside the country. Consequently, it contended that Indian courts lacked territorial jurisdiction.
Key Findings of the Court
Territorial Jurisdiction
Justice Amit Bansal rejected OpenAI's preliminary objection regarding maintainability and held that the Delhi High Court possessed territorial jurisdiction to adjudicate the dispute. The ruling confirms that Indian courts may entertain disputes involving global AI platforms where the alleged effects of the disputed activities are felt within India.
Fair Dealing and AI Training
A key aspect of the interim order concerns the application of the fair dealing exception under Section 52(1)(a) of the Copyright Act, 1957.
The Court observed that the storage and use of ANI's copyrighted material for training ChatGPT, at least at the interim stage, could fall within the scope of the statutory fair dealing exception and therefore would not constitute copyright infringement under Section 51 of the Act. While this finding is provisional and subject to final adjudication, it represents one of the earliest judicial considerations in India of whether the use of copyrighted works for AI training may be protected under existing copyright exceptions.
No Prima Facie Evidence of Copyright Infringement
The Court further held that ANI had failed to establish that ChatGPT reproduced its copyrighted content in a manner amounting to infringement.
Particular emphasis was placed on the operation of Retrieval-Augmented Generation (RAG), with the Court observing that the outputs generated by ChatGPT were not substantially similar to ANI's original reports. The Court also found no prima facie evidence that the model had "memorised" or reproduced ANI's copyrighted works.
Accordingly, the Court concluded that ANI had failed to establish a prima facie case warranting interim injunctive relief.
Balance of Convenience and Public Interest
In refusing to grant an interim injunction, the Court also considered broader public interest considerations.
Recognising the growing significance of AI technologies, the Court observed that data serves as the essential input for training LLMs and acknowledged the transformative impact of AI on access to information. It held that restraining ChatGPT at the interim stage could adversely affect technological innovation and the broader public interest, particularly before a full examination of the merits of the dispute.
Legal Significance
Although the decision is confined to an interim application and does not finally determine the parties' rights, it carries considerable significance for India's evolving AI jurisprudence.
The order indicates a judicial willingness to interpret existing copyright principles in the context of emerging AI technologies while balancing the competing interests of copyright owners, technological innovation and public access to information. It also demonstrates that Indian courts may be reluctant to impose broad interim restrictions on AI systems in the absence of clear evidence of copyright infringement.
Importantly, the Court did not conclusively determine whether AI training on copyrighted material is lawful under Indian copyright law. Those issues remain open for detailed examination during trial, and the final judgment may ultimately shape the contours of AI-related copyright protection in India, an issue already under review by the Ministry of Commerce's expert committee on AI and copyright.
Implications for Businesses
For AI developers, technology companies and organisations deploying generative AI, the order provides cautious optimism that Indian courts may allow innovation to continue while substantive legal issues are adjudicated.
For copyright owners and content creators, however, the decision reinforces that successful enforcement actions against AI developers are likely to require robust technical and evidentiary material demonstrating actual reproduction or infringement, rather than speculative assertions regarding model training.
As AI adoption accelerates across industries, this litigation is expected to play an important role in defining how Indian copyright law accommodates technological advancement while safeguarding the legitimate interests of content creators.
Looking Ahead
The Delhi High Court's interim order represents an important milestone in India's developing AI and copyright jurisprudence. While the merits of the dispute remain to be decided, the judgment offers valuable early insight into the legal framework that may govern AI training, copyright exceptions and innovation in India.
With AI regulation continuing to evolve globally, the final outcome of ANI Media Pvt. Ltd. v. OpenAI Inc. is likely to be closely watched by technology companies, media organizations, policymakers and legal practitioners alike, as it may influence the future relationship between intellectual property rights and artificial intelligence in India.
For more insights on data privacy and AI-related legal matters, feel free to reach out to our Partner, Dhruv Kaushal. He is a Certified Information Privacy Professional (Europe) [CIPP(E)] accredited by the International Association of Privacy Professionals (IAPP), demonstrating his expertise in global privacy and data protection frameworks, with a particular focus on data protection, privacy, artificial intelligence, intermediary regulations, and emerging technology laws.
Frequently Asked Questions
1. Did the Delhi High Court rule that AI training on copyrighted content is legal in India?
No. The Court's order is only an interim ruling on ANI's request for an injunction. It held that, prima facie, OpenAI's use of ANI's content for training fell within the fair dealing exception, but this is a preliminary observation, not a final determination. The main suit is still pending, and the question of whether AI training on copyrighted works is lawful under Indian law remains open for trial.
2. Why did the Court refuse to grant ANI an interim injunction against OpenAI?
The Court found that ANI failed to establish a prima facie case of infringement. It noted that ChatGPT's outputs, generated through Retrieval-Augmented Generation (RAG), were not substantially similar to ANI's original reports, and there was no evidence that the model had memorized or reproduced ANI's copyrighted content.
3. Does this ruling mean Indian courts have jurisdiction over foreign AI companies like OpenAI?
Yes, on this point the Court was clear. Justice Amit Bansal rejected OpenAI's objection that Indian courts lacked territorial jurisdiction, holding that Indian courts can hear disputes involving global AI platforms where the effects of the alleged conduct are felt in India, even if training occurs on servers located abroad.
4. What does this order mean for other content creators and publishers considering action against AI companies?
The order signals that Indian courts may be cautious about granting broad interim restrictions on AI systems without concrete evidence of infringement. Publishers and content owners will likely need robust technical and evidentiary material showing actual reproduction or memorisation, rather than relying on general allegations about how AI models are trained.