The Delhi High Court has denied ANI’s request to stop OpenAI from using the news agency’s works in model training while their copyright case proceeds. Justice Amit Bansal found, on a prima facie review, that storing the works to train ChatGPT’s underlying models fell within India’s fair-dealing exception for private or personal use, including research.
The 135-page order is the first detailed attempt by an Indian court to fit large-language-model training into the Copyright Act. It is not a final declaration that OpenAI prevailed on the merits, much less that every training practice is lawful.
The court separated ingestion from output
Reuters reported that ANI had not shown ChatGPT memorized or regurgitated its reports in responses to users. The court also found that the outputs ANI put forward, including responses generated with retrieval-augmented generation, were not substantially similar to its protected works. Copying material into a training process and reproducing protected expression in an output were therefore related but distinct questions.
That gives AI developers a useful litigation map under Indian law, but not a universal safe harbor. A claimant seeking similar interim relief will need evidence about memorization, substantial similarity or market effects rather than relying only on proof that copyrighted files entered a corpus.
Interim relief is not the final merits decision
The procedural posture should restrain the victory laps. ANI sought an interim injunction; the main suit remains pending. Its complaint, filed in November 2024, also alleged that ChatGPT generated fabricated stories and attributed them to ANI. The order’s copyright findings do not finally dispose of that reputational allegation or prevent a fuller factual record from changing the merits analysis.
The order also leaned on public interest. According to The Economic Times’ account, the court said an injunction at this stage could harm AI development and millions of Indian users. It nevertheless rejected OpenAI’s territorial objection and held that the Delhi court could hear the suit, a procedural win for ANI. Publishers will still contest the broader idea that the statutory reference to research can cover commercial model training.
For licensing teams, the immediate consequence is bargaining asymmetry, not legal certainty. OpenAI can point to a favorable Indian interim ruling; publishers can point to its preliminary posture and to different statutes in the United States, Canada and elsewhere. The next decisive evidence in this case is likely to come from model behavior and corpus records, not another abstract argument about whether training is transformative.
Sources: Reuters; SpicyIP; The Economic Times.
