The Delhi High Court's (“DHC/ the Court”) interim decision in ANI Media Pvt. Ltd. (“ANI”) v. OpenAI Opco LLC[1] (“OpenAI”) marks one of the first substantive judicial examinations in India of the intersection between copyright law and generative artificial intelligence (“AI”). At a time when courts across jurisdictions are grappling with whether the use of copyright-protected material for training large language models ("LLMs") constitutes copyright infringement, the DHC adopted a technologically informed and innovation-oriented interpretation of India's fair dealing doctrine. The judgment is particularly significant because India currently lacks a specific Text and Data Mining ("TDM") exception of the kind adopted in the European Union, Japan and Singapore; however, the Court has, at a prima facie level, interpreted Section 52(1)(a) of the (Indian) Copyright Act, 1957 (“the Act”) in a manner that partially addresses the absence of an express statutory TDM exception. While the observations are not final, they indicate a judicial willingness to interpret existing copyright principles in light of evolving AI technologies. It also aligns with emerging jurisprudence in the United States (“US”) and legislative developments globally, while simultaneously raising important questions regarding the continued suitability of the Department for Promotion of Industry and Internal Trade's ("DPIIT") proposed "One Nation, One License, One Payment" framework for AI training.

Introduction

The intersection of copyright law and AI represents one of the most important legal issues in the age of rampant technological advancements. Generative AI systems and their underlying LLMs, that are computed to serve as tools for content creation, research, brainstorming and idea generation, language translation, learning and education, creative writing, summarising or paraphrasing, require vast datasets to identify statistical patterns, linguistic relationships and semantic structures. From a copyright perspective, the central legal question is whether the ingestion and computational analysis of copyright-protected material for AI training constitutes an exercise of the copyright owner's exclusive rights or whether it qualifies as a permissible use under the Act as a transformative, research-oriented activity.

The Court's Key Observations

The DHC addressed novel legal challenges posed by generative AI’s use of freely available information, specifically examining key issues involving OpenAI’s storage and use of ANI’s copyright-protected news content to train the underlying LLM, and its generation of user responses that could potentially constitute copyright infringement under the Act. Further, the DHC determined whether the use of copyright-protected data for AI training qualifies as "fair dealing" under Section 52(1)(a) of the Act and whether Indian courts retain jurisdiction over the dispute given OpenAI's servers are in the United States.

The Court preliminarily divided the causes of action into two parts i.e., the training claim concerning the training of OpenAI’s LLMs using ANI’s copyright-protected content, and the output claim involving similarity between ANI’s copyright-protected content and OpenAI’s outputs to prompts – both of which ANI alleged constitute an infringement of the Act. Thereafter, the Court proceeded to evaluate these critical intersections of technology and intellectual property rights by framing the below major issues.

I. Territorial Jurisdiction of the Court and Applicability of the Act

The Court on a prima facie basis, held that it had territorial jurisdiction under Section 62(2) of the Act and Section 20 of the Civil Procedure Code, 1908 ("CPC") over both the output and training claims, and rejected OpenAI's contention that the location of its US servers excluded the application of the Act. The Court observed that accepting OpenAI's jurisdictional objection would permit entities to avoid the application of Indian copyright law merely by locating their servers outside India, notwithstanding that the allegedly infringing acts bear a substantial nexus with India. It highlighted that storage on US servers is merely the terminal step in a chain beginning with the access and transmission of copyright-protected works from India, and severing this chain would allow infringers to evade domestic law by routing operations offshore.

II. ANI failed to establish substantial similarity between its content and OpenAI’s output

a. No Evidence of Memorisation or Regurgitation

The Court, on the basis of OpenAI’s training predating the articles cited by ANI, prima facie concluded that ANI had failed to establish that ChatGPT (OpenAI's AI tool) memorised and regurgitated ANI's works. Questions concerning memorisation were treated as evidentiary matters requiring trial and expert examination.

b. Outputs Are Not Substantially Similar to Inputs

The Court further held that the outputs generated by ChatGPT did not prima facie infringe ANI’s copyright as they were not substantially similar to ANI’s input material.

The Court thus drew an important distinction between the use of copyright-protected works during AI model development and the assessment of allegedly infringing outputs generated by the trained model. This distinction is likely to assume increasing significance as future disputes focus on whether particular AI-generated outputs reproduce protected expression rather than whether the training process itself infringes copyright.

c. Publisher Opt-Outs Mechanism

The Court highlighted that the publishers retain the ability to use opt-out mechanisms to prevent web crawlers from replicating content or scraping platforms for retrieval-augmented generation (RAG) and search capabilities. Although this was not treated as determinative of the copyright analysis, it formed part of the broader factual context considered by the Court.

III. Training of LLMs, falling within the exceptions to copyright infringement, cannot violate the Act

a. AI Training Falls Within Fair Dealing

The Court found that OpenAI's storage of ANI’s copyrighted material for training of its LLMs would prima facie fall under the fair dealing exception of Section 52(1)(a) of the Act.

This is likely to be one of the most significant aspects of the interim order from the perspective of both AI developers and copyright owners. Analysing the legislative history of Section 52(1)(a) of the Act, i.e., “Private or Personal Use, including Research”, the Court has prima facie reaffirmed that the LLMs with stored literary work meet the “Purpose Test”. Further, based on the three parameters i.e. Containment to model training, potential for market substitution and public interest, the Court found that the “Fairness Test” was also met. The Court recognised that AI training constitutes an internal computational process through which models learn patterns and relationships rather than reproduce copyright-protected works for public consumption. By characterising the activity as a form of research falling within Section 52(1)(a), the Court effectively adapted the existing copyright provision to an AI age technological context. Relying on the doctrine of ‘updating construction’, the Court held that research can no longer be restricted to humans; rather, it must be extended to AI in view of technological advancements, to give effect to the purpose of the Act.

Notably, the Court's reasoning appears to move Indian copyright jurisprudence towards a more functional assessment of fair dealing, focusing on the purpose, nature and economic impact of the use rather than on whether copyright-protected works are copied during the training process. Although the Indian fair dealing doctrine has traditionally been interpreted more narrowly than the US doctrine of fair use, the interim order suggests a greater willingness to interpret the statutory exception in a technologically contextual manner where the impugned activity does not substitute the original work.

b. Machine Learning as an Internal Research Process

The Court also emphasised that model training is "completely an internal process" and does not amount to making copyright-protected works available to third parties. The Court further noted that “research and learning” is no longer confined to humans and now extends to AI functioning at human direction for human benefits.

This distinction is important because it acknowledges a central feature of machine learning: training datasets are analysed to derive statistical representations and model weights rather than to distribute copyright-protected expression itself.

c. Commercial Training Protected

The Court has further interpreted that the defence under Section 52(1)(a) remains applicable even if the internal research and training are conducted for “commercial purposes”. This observation recognises that commercially valuable technological innovation may nevertheless constitute research for the purposes of copyright law.

d. Public Interest Favours AI Development

The Court's reliance on public interest is particularly noteworthy. Unlike conventional copyright disputes involving competing commercial interests, the Court recognised that AI model development has broader implications for technological advancement, research and India's aspirations to emerge as a global AI hub. This introduces an important public policy dimension into copyright adjudication involving AI systems. Perhaps the strongest policy statement contained in the order was the Court's observation that:

"Any interim injunction granted at this stage would be detrimental to the growth of AI and more particularly to the LLMs being developed in India."

The Court thereby expressly recognised the broader societal benefits associated with AI innovation and India's ambitions to emerge as a global AI hub.

Comparative Global Jurisprudence

European Union: The European Union continues to recognise the dual TDM exceptions under Articles 3 and 4 of the DSM Copyright Directive. Recent judicial developments, particularly the German Regional Court's decision in Kneschke v. LAION e.V.[2], have indicated a willingness to treat AI training datasets as falling within the TDM framework, subject to compliance with the Directive's conditions. At the policy level, however, the debate continues. The European Parliament's 2025 study on Generative AI and Copyright questions whether Article 4 was originally intended to extend to foundation model training and suggests that legislative clarification may ultimately be required.

United Kingdom (UK): The UK has recently reconsidered proposals to introduce a broader commercial TDM exception. Following extensive stakeholder consultations, the Government concluded that additional evidence is required before undertaking significant legislative reform. Accordingly, the existing position remains unchanged, with the statutory TDM exception under the Copyright, Designs and Patents Act 1988 (“CDPA”) continuing to apply only to non-commercial research. In Getty Images v. Stability AI[3], the UK High Court held that AI model weights do not constitute “copies” of the copyrighted works under the CDPA, as they contain statistically trained parameters rather than stored or recognisable reproductions of the underlying works; accordingly, their importation, possession or distribution in the UK did not amount to secondary copyright infringement. The UK's evolving approach demonstrates a measured attempt to balance technological innovation with the protection of creative industries while allowing the courts to continue developing the law[4].

US: Recent US decisions have increasingly distinguished between the acquisition of copyright-protected material, the computational process of AI training, and the generation of allegedly infringing outputs. In Bartz v. Anthropic[5], the court recognised that training AI models using lawfully acquired works may constitute a highly transformative use because the models learn statistical relationships rather than expressive content. Similarly, in Kadrey v. Meta[6] the Court has ruled in favour of fair use (the Court focused on issues such as market substitution, actual copying and evidentiary proof rather than treating AI training itself as inherently infringing). In Authors Guild v. Google Inc. and Authors Guild v. HathiTrust[7] courts upheld mass digitisation projects because the uses were transformative and did not substitute for the original works. Although not AI cases, the Authors Guild decisions established the principle that copying entire works for computational analysis may nevertheless constitute fair use. These developments broadly align with the DHC’s distinction between AI model training and the assessment of AI-generated outputs.

Japan: Japan continues to maintain one of the most permissive statutory frameworks for AI development. Article 30-4 of the Japanese Copyright Act permits the use of copyright-protected works for information analysis, including AI training, irrespective of whether the activity is commercial. Recent guidance[8] has clarified that the exception does not extend to outputs that reproduce protected expression or other uses that conflict with the legitimate interests of rightsholders. Japan therefore continues to distinguish between computational learning and infringing exploitation while preserving a broad statutory exception for AI training.

Singapore: Singapore continues to recognise one of the clearest statutory exceptions for computational data analysis. The (Singapore) Copyright Act expressly permits computational data analysis, including machine learning and AI training, provided the user has lawful access to the underlying material. The legislation does not distinguish between commercial and non-commercial AI development, thereby providing a comparatively stable legal framework[9] while continuing to require lawful access to copyright-protected works.

The DHC's prima facie reasoning therefore reflects an emerging international trend of distinguishing AI training from infringing exploitation of copyright-protected works, while simultaneously demonstrating that jurisdictions continue to adopt different legislative and judicial approaches to balancing innovation with the interests of copyright owners.

The Judgment as India's De Facto TDM Exception

Perhaps the most overlooked aspect of the decision is its significance in filling a legislative gap.

Unlike the European Union's DSM Copyright Directive[10], which expressly creates TDM exceptions under Articles 3 and 4, India's Copyright Act contains no equivalent provision. Similarly, Japan and Singapore have enacted specific exceptions recognising computational analysis and machine learning activities.

In the absence of such legislation, the DHC effectively used and purposively extended Section 52(1)(a)'s fair dealing framework to perform the same function.

The Court recognised that machine learning is fundamentally different from expressive consumption. Instead of reading fair dealing through the lens of traditional copying, the Court interpreted it through the lens of computational analysis and AI research. This is arguably the most progressive feature of the judgment.

Accordingly, ANI v. OpenAI can be viewed as India's first judicial recognition that fair dealing must evolve to accommodate machine learning technologies in the absence of a dedicated TDM exception.

Objectives of India's AI Governance Framework

More broadly, the judgment reflects a technology-neutral approach to statutory interpretation. Rather than waiting for legislative intervention, the Court interpreted existing copyright principles in a manner capable of accommodating technological developments while continuing to preserve the fundamental objectives of copyright protection. This helps to align with the objectives of India's evolving AI governance[11] framework by fostering innovation and enabling provisions to protect the rights of rightsholders.

Unlike the European Union's AI Act[12] model, India's governance strategy has largely avoided ex ante restrictions and instead focused on enabling innovation while developing safeguards through responsible AI principles. The DHC’s refusal to impose a broad injunction fits squarely within this philosophy. By recognising AI training as a socially beneficial activity and applying copyright exceptions accordingly, the Court reinforced India's innovation-first policy orientation.

The DPIIT Working Paper: A Proposal Likely to Prompt Further Policy Reconsideration

The DHC's decision also has important implications for the DPIIT Working Paper[13], "One Nation, One License, One Payment", released in December 2025. The Working Paper rejects a pure TDM exception and instead proposes a mandatory blanket licensing regime administered by a centralised Copyright Royalties Collective for AI Training ("CRCAT"). The DPIIT Working Paper proposes a Hybrid Model to ensure that copyright-protected works are automatically available for training of LLMs and owners of such works are fairly compensated through centralised CRCAT.

The proposed royalty mechanism raises practical concerns, particularly as the foundation models are trained on billions of data points from millions of sources. Determining the relative contribution of specific works and administering royalty distributions presents substantial complexity and has various other practical challenges[14]. The report has been opposed by the industry associations[15] as well as the publishers[16]. At a strategic level, the proposal also risks putting India at odds with emerging global approaches. The European Union, Japan, Singapore, and recent US jurisprudence increasingly recognise that computational learning should be treated differently from copyright exploitation. Introducing a complex royalty-based architecture where major jurisdictions are moving toward broader TDM and fair-use style approaches could create regulatory uncertainty and increase compliance burdens for Indian innovators.

The more sustainable long-term solution may be a carefully designed statutory TDM exception that balances innovation, creator interests, transparency obligations, and safeguards against market substitution. Such an approach would provide legal certainty while avoiding the economic and administrative complexities inherent in a compulsory licensing regime.

Conclusion

The DHC's interim order in ANI Media Pvt. Ltd. v. OpenAI is far more than a procedural ruling denying interim relief. It constitutes a sophisticated attempt to reconcile copyright law with machine learning technologies and establish a doctrinal pathway for AI innovation within India's existing legal framework.

By holding that AI training may constitute fair dealing, by emphasising the transformational nature of machine learning, by rejecting unsupported allegations of memorisation, and by recognising the public interest in AI development, the Court has effectively applied through judicial interpretation what Parliament has not yet enacted through legislation: a functional equivalent of a TDM exception.

Looking ahead, the interim order is likely to play a significant role in informing future judicial decisions as well as ongoing policy discussions surrounding the regulation of AI training datasets. Whether India ultimately chooses to introduce an express statutory TDM exception, retain reliance on judicial interpretation of fair dealing, or adopt a licensing-based framework remains a question for future legislative consideration. The outcome of the final adjudication in ANI v. OpenAI will therefore be closely watched by AI developers, copyright owners and policymakers alike. As global copyright law increasingly recognises the distinction between computational learning and expressive exploitation, India's future policy framework may be better served by a clear statutory TDM exception rather than a complex royalty-driven licensing architecture.

_________________________________________________________________________

[1] 2026 SCC OnLine Del 5291; Available at: https://delhihighcourt.nic.in/app/showFileJudgment/ABL24072026SC10282024_171649.pdf

[2] District Court of Hamburg, Robert Kneschke v. LAION e.V., Case No. 310 O 227/23; Available at: https://www.wipo.int/wipolex/en/text/592042

[3] Getty Images (US) Inc & Ors v. Stability AI Limited [2025] EWHC 2863 (Ch). Getty Images v Stability AI [2025] EWHC 2863 (Ch) Available at: https://www.judiciary.uk/wpcontent/uploads/2025/11/Getty-Images-v-Stability-AI.pdf

[4]UK Government, Report on Copyright and Artificial Intelligence (2026); Available at: https://www.gov.uk/government/publications/report-and-impact-assessment-on-copyright-and-artificial-intelligence/report-on-copyright-and-artificial-intelligence

[5] 787 F. Supp. 3d 1007 (N.D. Cal. 2025); Available at: Bartz et al v. Anthropic PBC, No. 3:2024cv05417 - Document 437 (N.D. Cal. 2025) :: Justia

[6] Kadrey et al v. Meta Platforms, Inc., No. 3:2023cv03417 - Document 598 (N.D. Cal. 2025); Available at: https://storage.courtlistener.com/recap/gov.uscourts.cand.415175/gov.uscourts.cand.415175.1.0_3.pdf

[7] 804 F.3d 202 (2d Cir. 2015) and 755 F.3d 87 (2d Cir. 2014); Available at: Authors Guild v. Google, Inc., No. 13-4829 (2d Cir. 2015) :: Justia

[8] Available at: AI White Paper 2024.pdf

[9] Available at: Model-AI-Governance-Framework-for-Generative-AI-May-2024-1-1.pdf

[10] European Union, Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on Copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC, (17 May 2019), Available at: DIRECTIVE (EU) 2019/ 790 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL - of 17 April 2019 - on copyright and related rights in the Digital Single Market and amending Directives 96/ 9/ EC and 2001/ 29/ EC

[11] Ministry of Electronics and Information Technology, India AI Governance Guidelines, Enabling Safe and Trusted AI Innovation, Available at: Final_Version_01.

[12] European Union, Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act), (12 July 2024), Available at: Regulation - EU - 2024/1689 - EN - EUR-Lex.

[13] Department for Promotion of Industry and Internal Trade, Ministry of Commerce and Industry, One Nation One License One Payment- Balancing AI Innovation and Copyright, (December 2025), Available at: ff266bbeed10c48e3479c941484f3525.pdf

[14] AI copyright, dead on arrival?; Available at: https://economictimes.indiatimes.com/opinion/et-commentary/ai-copyright-dead-on-arrival/articleshow/126082127.cms?

[15] Nasscom pushes back against DPIIT plan; Available at: https://www.hindustantimes.com/india-news/nasscom-pushes-back-against-dpiit-plan-101765436303176.html

[16] Publishers oppose DPIIT’s AI licensing paper, warn of risks to news industry and IP rights; Available at: Publishers oppose DPIIT’s AI licensing paper, warn of risks to news industry and IP rights - Storyboard18

 _______________________________________

The article has been authored by Partner, Deepak Singh along with Partner, Avinash Amarnath and Associates, Shivangi Mugdha and Lalitha Durvasula. The views expressed here are their own.

In case of feedback, suggestions, or queries, please reach out to the C&M team on

[email protected]