Please click here to download the Prism as a PDF.
Delhi High Court rules that the use of publicly available news content for training prima facie constitutes fair dealing
In ANI Media Private Limited vs. OpenAI OpCo LLC[1], the Delhi High Court (“Delhi HC”) has delivered the first substantive Indian ruling on the intersection of Artificial Intelligence (“AI”) training and copyright law. Dismissing the interim injunction application filed by ANI Media Private Limited (“ANI”) against OpenAI OpCo LLC (“OpenAI”), the Delhi HC, on a prima facie basis, held that storing freely available copyrighted news content to train large language models constitutes fair dealing under Section 52(1)(a)[2] of the Copyright Act, 1957 (“Copyright Act”). The ruling interprets India’s fair dealing framework expansively, recognises the public interest in AI development. It signals that using publicly accessible material for model training may not, by itself, infringe copyright, a development with significant implications for AI developers and content owners operating in India.
Brief facts
ANI, a news agency, filed a suit against OpenAI in November 2024[3] alleging copyright infringement on two counts. First, ANI alleged that OpenAI copied and stored ANI’s news articles and interviews to train the large language models underlying ChatGPT. Second, it alleged that OpenAI’s ChatGPT reproduced ANI’s works in responses generated for users.
ANI sought an interim injunction restraining OpenAI from storing, publishing, reproducing or otherwise using ANI’s works through ChatGPT, and a direction requiring OpenAI to disable ChatGPT’s access to works published by ANI or its subscribers. OpenAI denied infringement and stated that the relevant model training had ended in April 2022 for GPT-4 and April 2024 for GPT-4o. Open AI had blacklisted ANI’s website in October 2024 and later stated that it had stopped using ANI’s website for both training and retrieval-augmented generation (“RAG”)[4].
Issues
- Whether the storage by the defendants of plaintiff data (which is in the nature of news and is claimed be protected under the Copyright Act) for training the software i.e., ChatGPT, would amount to infringement of plaintiff’s copyright?
- Whether the use by the defendants of plaintiff’s copyrighted data in order to generate responses for its users, would amount to infringement of the plaintiff’s copyright?
- Whether the defendants’ use of plaintiff’s copyrighted data qualifies as ‘fair use’ in terms of Section 52 of the Copyright Act?
- Whether the courts in India have jurisdiction to entertain the present lawsuit considering that the servers of the defendants are located in the United States of America?
Findings and analysis
The Delhi HC first considered Issue 4 pertaining to jurisdiction. Thereafter the Delhi HC opined on Issue 2 and concluded with its ruling in relation to Issues 1 and 3 which, the Delhi HC stated are intertwined.
Issue No. 4: Whether the courts in India have jurisdiction to entertain the present lawsuit considering that the servers of the defendants are located in the United States of America
The Delhi HC held that it had territorial jurisdiction under Section 62(2) [5] of the Copyright Act because ANI has its principal place of business and registered office in Delhi. Jurisdiction also arose under Section 20 of the Code of Civil Procedure, 1908[6] because OpenAI offered and targeted its services to users and paid subscribers in India, including Delhi, and the outputs relied on by ANI were generated and accessed within Delhi.
The Delhi HC also opined that the location of Open AI’s servers in the United States does not defeat jurisdiction. It described the storage on overseas servers as the terminal step in a chain of events that began with access to works in India and their transmission abroad. It held that separating only the last step could allow a person to evade Indian copyright law by moving the terminal link to a server outside India. The training and output claims were also connected because the outputs were based on the trained model. The Delhi HC therefore held, on a prima facie basis, that it has jurisdiction.
Issue No. 2: Whether the use by the defendants of plaintiff’s copyrighted data in order to generate responses for its users, would amount to infringement of the plaintiff’s copyright
The Delhi HC accepted, on a prima facie basis, that ANI owned copyright in original literary works published on its website under its agreements with its professionals. It also held that public availability does not remove copyright protection. However, copyright in news extends only to the original form and manner in which facts are expressed. It does not extend to the underlying facts. Since news primarily reports actual events and facts, the Delhi HC held that the threshold for proving substantial similarity in protected expression is higher.
ANI relied on 9 (nine) examples to allege that ChatGPT had memorised and reproduced its works. The Delhi HC noted that all 9 (nine) examples cited by ANI were published after the training cut-off dates of the ChatGPT models used by OpenAI. The examples could therefore not establish memorisation through model training. The Delhi HC considered that the responses were more likely generated by retrieving live information through RAG. It also recorded that whether RAG outputs could infringe copyright had not been pleaded as a distinct claim. For the outputs placed before it, the Delhi HC found no substantial similarity between ChatGPT’s responses and ANI’s articles.
The Delhi HC applied the test in R.G. Anand vs. Deluxe Films[7] and compared each work, rather than selected extracts. In the example concerning the statement made by Neeraj Chopra’s mother, the Delhi HC found that the first response was “not a substantial reproduction or nearly exact copy” and that ChatGPT had added commentary which bore ‘its own expression’ and was ‘quite dissimilar’ to ANI’s expression. The second response followed an adversarial prompt in which ANI asked ChatGPT to state ‘exactly’ what Neeraj Chopra’s mother had said. The Delhi HC found that the only similarity related to part of a quotation from the interview. Applying Section 17(cc) of the Copyright Act[8], the Delhi HC held, on a prima facie basis, that Neeraj Chopra’s mother was the first owner of copyright in the quotation. It further held that the right to translate the quotation vested with her under the Copyright Act, and ANI had not produced any evidence of an assignment in its favour. The remainder of ChatGPT’s response contained its own explanatory text. The Delhi HC concluded that ANI had not shown that ChatGPT permanently stored, memorised or regurgitated ANI’s works, or that the responses were a substantial reproduction of those works. and prima facie held that no case of infringement was made on the output claim.
Issue No. 1: Whether the storage by the defendants of plaintiff’s data (which is in the nature of news and is claimed to be protected under the Copyright Act) for training its software i.e., ChatGPT, would amount to infringement of plaintiff’s copyright
The Delhi HC held that Section 14(a)(i) of the Copyright Act gives the copyright owner the exclusive right to reproduce a literary work in any material form, including storing it in any medium through electronic means. The provision does not distinguish between temporary and permanent storage, and the purpose or intention behind storage is not relevant at the infringement stage. Open AI’s temporary storage of ANI’s works for training therefore falls within the reproduction right and would ordinarily attract Section 51 of the Copyright Act[9].
However, Section 14 is expressly subject to the other provisions of the Copyright Act. The Delhi HC held that Section 52 of the Copyright Act is an integral part of the statutory scheme and independently defines user rights in copyrighted works. An act that satisfies Section 52 of the Copyright Act does not amount to infringement. Section 52 of the Copyright Act should therefore receive a broad and liberal interpretation that balances the rights of copyright owners with creativity, research and the dissemination of knowledge. The Delhi HC consequently considered Issue Nos. 1 and 3 together.
Issue No. 3: Whether the defendants’ use of plaintiff’s copyrighted data qualifies as ‘fair use’ in terms of Section 52 of the Copyright Act
The Delhi HC applied a 2 (two)-step inquiry under Section 52(1)(a) of the Copyright Act; first, whether the use was for one of the purposes listed in that provision (the ‘purpose test’); and second, whether the dealing was fair (the ‘fairness test’).
Purpose test
The Delhi HC held that Section 52(1)(a) of the Copyright Act pertaining to fair dealing is not limited to non-commercial use. The legislature, while drafting the Copyright Act has expressly used the words non-commercial in certain other clauses of Section 52 of the Copyright Act, but not in Section 52(1)(a) of the Copyright Act. A commercial purpose may be relevant to fairness but does not in itself prevent a person or company from relying on the fair dealing provision.
The Delhi HC held that the words “not itself being an infringing copy” set out in the explanation to Section 52(1)(a) of the Copyright Act qualify incidental storage of a computer programme, and do not impose a general condition on every work stored electronically. It also noted that ANI did not allege that OpenAI obtained the works from an unauthorised source or broke through a paywall. The works were freely available on ANI’s website.
The Delhi HC held that the term ‘private’ as set out in Section 52(1)(a)(i) is wider than ‘personal’ and can include a group, institution or private company. The training data was kept in a closed environment, was accessible only to the model, and was not available to the public for access or download. The Delhi HC also interpreted the term ‘research’ broadly and opined that research is not confined to acts performed by human beings and can extend to machine learning. The training process, therefore, the Delhi HC view, fulfilled the purpose test.
Fairness test
The Delhi HC noted that Indian courts have not adopted one uniform test for fair dealing and that the 4 (four)-factor tests under the United States law does not directly apply in India. It formulated 3 (three) factors suited to the case, broadly in line with Article 9 of the Berne Convention[10]; whether the use was limited to training; whether it created economic competition, market substitution or actual or potential harm to ANI; and whether the functions performed by ChatGPT served the public interest.
First, the Delhi HC stated that ANI had not established memorisation, regurgitation or communication of exact copies, and no instance showed that OpenAI used ANI’s works for a purpose other than training. Second, ANI and ChatGPT performed different functions. ANI collected, verified and syndicated news, while ChatGPT was a general-purpose tool used for analysis, text generation, education, translation, research and other tasks. ANI produced no evidence of loss of subscribers, advertising revenue or market share. The Delhi HC therefore found no market substitution or actual or potential damage.
Third, the Delhi HC placed weight on the public benefits of trained large language models, including their use in education, scientific research, software development, translation, accessibility and the wider dissemination of knowledge. It held that the use served the public interest. Since both the purpose test and the fairness test were satisfied, the Delhi HC in its preliminary ruling opined that OpenAI’s storage of ANI’s literary works for training fell within Section 52(1)(a) and did not amount to infringement at the interim stage.
Balance of convenience and irreparable injury
The Delhi HC found no material showing that OpenAI’s activities had caused ANI to lose subscribers or revenue. Further, the Delhi HC noted that ANI had offered OpenAI a license for its content for USD 7,500,000 (US Dollars seven million five hundred thousand), which showed that any proven loss could be quantified and compensated by damages. By contrast, an injunction requiring deletion or restraint could materially affect OpenAI’s operations and the public. The Delhi HC also considered that access to data is important for the development of large language models and that an injunction could affect AI development in India. It held that the balance of convenience and irreparable injury tests were against ANI and dismissed the injunction application.
At this juncture it is pertinent to note that all the aforesaid observations of the Delhi HC were expressly limited to interim injunction application and are not the final decision in the suit which the Delhi HC has not yet reached a verdict on.
Conclusion
The judgment is the first substantive Indian ruling on the use of copyrighted news content for AI training. It provides interim support for AI developers where the material is freely available and not obtained from unauthorised sources, copies remain within a closed training environment, outputs do not reproduce protected expression, and there is no evidence of market substitution. However, the ruling is not a general license to use all copyrighted material for AI training, and different facts, such as the use of pirated datasets, circumvention of access controls, persistent storage for purposes other than training, or proof that an AI service substitutes the copyright owner’s product, may lead to a different result.
Its reasoning is closely linked to the facts presented before the Delhi HC , including the public availability of ANI’s works, the absence of paywall circumvention or unauthorised copies, the lack of substantial reproduction or regurgitation, and ANI’s failure to establish loss of subscribers, revenue or market share. Different facts, such as the use of pirated datasets, circumvention of access controls, persistent storage of works for purposes other than training, repeated memorisation, substantially similar outputs, or proof that an AI service substitutes the copyright owner’s product, may lead to a different result.
While the Delhi HC’s preliminary ruling in relation to the injunction application is in OpenAI’s favour, in the Delhi HC’s own words, the observations set out in the interim ruling will have no bearing on the final outcome of the suit.
This Prism has been prepared by:
|
Sajai Singh |
Sankalp Inuganti |
Adarsh Tripathi |
For more details, please contact km@jsalaw.com.
[1] DHC: 5900, I.A. 45300/2024 in CS(COMM) 1028/2024 (decided on July 24, 2026)
[2] Section 52 of the Copyright Act identifies specific acts that do not constitute copyright infringement. These include fair dealing for purposes such as private or personal use, research, criticism, review and reporting current events, along with certain other permitted uses listed in the provision.
[3] CS(COMM) 1028/2024
[4] Where an LLM accesses information that was not part of its training data, it may do so through retrieval-augmented generation, or RAG.
[5] Section 62(2) of the Copyright Act provides an additional forum for instituting copyright proceedings. A plaintiff may sue before the district court within whose jurisdiction the plaintiff actually and voluntarily resides, carries on business, or personally works for gain, notwithstanding the ordinary jurisdiction rules under the Code of Civil Procedure, 1908.
[6] Section 20 of the Code of Civil Procedure, 1908 sets out the ordinary rules of territorial jurisdiction. A suit may generally be filed where the defendant resides, carries on business or works for gain, or where the cause of action arises wholly or in part.
[7] R.G. Anand v. Deluxe Films, (1978) 4 SCC 118.
[8] Section 17(cc) of the Copyright Act provides that the person who delivers an address or speech in public is the first owner of copyright in relation to such an address or speech, unless it is delivered on behalf of another person, in which case that other person is the first owner.
[9] Section 51, Copyright Act, 1957 lays down instances when a copyright is considered to be infringed.
[10] Article 9 of the Berne Convention gives authors the exclusive right to authorise reproduction of their literary and artistic works in any manner or form. It permits member countries to create exceptions only in certain special cases where the reproduction does not conflict with the normal exploitation of the work or unreasonably prejudice the author’s legitimate interests.
