Read Time: 07 minutes

Synopsis

The Delhi High Court, on Friday, heard the submissions of Amicus Curiae Arul George Scaria on the issue: whether the storage of ANI’s copyrighted material for AI training amounted to infringement. The bench of Justice Amit Bansal adjourned the matter for March 10, 2025. 

Diving into the intricate provisions of the Copyright Act, Amicus Curiae Scaria highlighted key sections such as 16, 30, and 40, before focusing on the crucial Section 17, which defines the first owner of copyright. 

The Court engaged in a pointed discussion on whether ANI, as a news agency, could claim first ownership, given that its journalists might not be direct employees but contributors. The Amicus emphasized that an employer-employee relationship was essential to establish ownership under Section 17(b), which explicitly excluded certain literary works from its scope.

The discussion then turned to the mechanics of AI training. Amicus Curiae Scaria explained the technical nuances of how language models are trained. He described the process of tokenization, where collected data—including potentially copyrighted content—was converted into numerical tokens. He clarified that these models did not retain or reproduce source material verbatim but rather identified statistical relationships between words.

Justice Bansal, intrigued by the explanation, sought clarification on whether the process involved a one-time access to copyrighted material or if repeated reference was necessary. Amicus Curiae Scaria responded that once tokenized, there was no need to return to the original content, thus posing an important question—whether such non-expressive use could even constitute infringement.

The Court acknowledged that proper attribution was a relevant concern but pressed further on whether such use was permissible under the fair use exceptions of Section 51.

Amicus Curiae Scaria then laid out the broader implications, noting that India’s copyright framework had built-in provisions for learning-based use. He referenced Section 52, which permitted certain non-expressive uses and fair dealing exceptions. While he conceded that news reporting had elements of copyright protection, he argued that facts themselves were not copyrighted—only their specific narration could be.

The bench posed a crucial hypothetical: if AI companies were required to pay licensing fees for every dataset they trained on, would such a model even be viable? Amicus Curiae Scaria stressed that the court must balance the interests of copyright holders and technological innovation, especially as large language models (LLMs) were still in their developmental phase. Furthermore, he pointed out that ANI already licensed its content to other entities, raising questions about whether an opt-out mechanism for AI training would suffice.

As the proceedings drew to a close, the discussion shifted to misinformation concerns. Amicus Curiae Scariacontended that restricting AI’s access to information could, paradoxically, worsen the problem. Instead, he advocated for greater transparency and access to mitigate the spread of false narratives.

The case was filed by ANI Media, alleging that OpenAI had infringed upon its copyright by utilizing its content without authorization. ANI asserted that ChatGPT used its material to train the AI software, which is then stored and provided access to ANI’s content for users. However, in the previous hearing, OpenAI contended that since the company neither operates in India nor are its servers located in India, therefore no cause of action arises in India. 

For Plaintiff: Advocate Sidhanth Kumar
For Respondent: Senior Advocate Akhil Sibal
Case Title: ANI Media Pvt Ltd v OpenAI Inc (CS COMM 1028/2024)