Delhi HC says learning AI is for ‘personal and domestic use’

The Delhi High Court refused to grant ANI, a news agency based in India (Asian News International), an interim injunction against OpenAI. In his judgment, Justice Amit Bansal held, for the first time in the history of the world, that the training of a language model constitutes ‘private or personal use’ within the meaning of copyright law in India.

1ANI's Evidence Fell Apart

He held that GPT-4 and GPT-4o were trained on pre-April 2022 and pre-April 2024 data respectively. ANI’s claim of copyright infringement was based on its own articles published in August–September 2024 (after the training dates of the respective models). Even with the help of custom-built adversarial prompts, it could not obtain any exact reproduction of text from its own articles through ChatGPT.

The court held that while copyright subsists in the manner of expression of facts in news articles, it does not cover the facts themselves. This weakens ANI’s argument.

2A Three-Part Test

To determine whether the use of the work was permissible for training purposes, the court considered three factors: (a) whether the use was restricted (only copies used for training and not shared publicly); (b) whether the use caused economic loss (AI companies and news agencies belong to distinct markets); and (c) whether the use benefits society (education, research, access to information, aiding the disabled). All these criteria were satisfied.

3A Legal Win That Doesn't Solve the Economic Problem

The court has clearly distinguished between the use of lawfully obtained data for model training, which is not the same as copying, and the impact of AI-driven search on media companies’ revenues, which is outside the purview of this case. As per Pew Research, CTRs decline from 15% to 8% if AI summaries are displayed. Users find what they want and leave. Winning in court won’t bring back readers.