Landmark Decision: Delhi High Court issues Injunction in ANI v. OpenAI Copyright Dispute
In a landmark decision for India’s intellectual property and technology ecosystem, the Delhi High Court refused to grant an interim injunction sought by news agency Asian News International (ANI) against OpenAI OpCo LLC. The order in ANI Media Pvt. Ltd. v. OpenAI OpCo LLC [CS(COMM) 1028/2024] represents the Indian judiciary’s first detailed judicial assessment on whether training Large Language Models (LLMs) on copyrighted works constitutes copyright infringement.
Key Legal Findings
1. Training LLMs as “Fair Dealing” Under Section 52(1)(a)
The central issue before the Court was whether scraping, storing, and using copyrighted articles to train ChatGPT constitutes infringement under Section 51 of the Copyright Act, 1957.
Justice Amit Bansal observed that storing data in a closed, private digital repository solely for training AI models qualifies prima facie under the statutory defense of “private or personal use, including research” pursuant to Section 52(1)(a)(i). The Court clarified that:
- The term “private” is not restricted to individual natural persons and can extend to corporate or private entities storing data in non-public environments.
- Commercial intent does not automatically bar a party from invoking fair dealing under Section 52(1)(a), as the statute does not contain an explicit non-commercial limitation for private research.
- Statutory interpretation must evolve dynamically to account for modern technological paradigms where “research” is conducted via machine learning algorithms rather than exclusively by human researchers.
2. Generative Outputs Do Not Amount to Substantial Reproduction
ANI argued that ChatGPT generated responses containing portions of its proprietary articles. The Court distinguished between factual information (unprotectable) and creative expression (protectable under the Fact/Expression Dichotomy).
It concluded that ChatGPT’s generative outputs—utilizing architecture such as Retrieval-Augmented Generation (RAG) did not display substantial literal or non-literal copying of ANI’s original expression, thereby failing to establish prima facie output infringement.
3. Balance of Convenience & Public Interest
In evaluating whether to grant interim relief, the Court emphasized that:
- Quantifiable Harm: ANI’s claims could be monetarily compensated, pointing out prior licensing negotiations.
- Public and Economic Interest: Restraining generative AI technologies at an interim stage would choke technological innovation in India and deprive millions of users of AI-assisted productivity tools.
Comparative Perspective: India vs. US and EU Frameworks
The Delhi High Court’s approach positions India as a distinct, technology-friendly jurisdiction compared to global legal frameworks:
| Jurisdiction | Legal Mechanism | General Judicial & Legislative Approach |
| India | Section 52(1)(a) Fair Dealing | Broad interpretation of “private use/research” covering automated machine training in closed systems. |
| United States | Section 107 Fair Use Doctrine | Multi-factor fair use analysis; ongoing litigation (e.g., NYT v. OpenAI) scrutinizes market substitution and web scraping. |
| European Union | EU AI Act & TDM Exception | Strict statutory Text and Data Mining (TDM) exceptions with explicit publisher opt-out mechanisms. |
What Lies Ahead?
While this interim order sets a persuasive benchmark favoring AI developers, it is not a final adjudication. The main suit is scheduled to proceed to full trial, where both parties will present evidence on text data mining, economic impact, and licensing models.
For copyright owners, digital publishers, and enterprise AI companies, this decision underlines the urgent need for clear contractual licensing terms, robust web-crawler management strategies (robots.txt), and structured AI compliance frameworks.
