Delhi High Court denies ANI injunction against OpenAI
India's landmark AI copyright case turned on evidence: ANI's cited articles postdated model training, and the judge treated LLM training as transformative.
The ruling
The Delhi High Court on July 27 declined to grant Asian News International a preliminary injunction against OpenAI, in the highest-profile AI copyright case brought in India. Justice Amit Bansal's interim order rejects ANI's request to restrain OpenAI from using its journalism in training and in ChatGPT outputs.
The evidentiary problem was decisive. Several articles ANI submitted as examples of copied work were published after the relevant model's training data cutoff, meaning they could not have been ingested — an inconsistency that undercut the broader infringement claim. Beyond that, the court found no demonstrated economic harm, reasoning that a news agency and an AI developer are not competitors in the same market. It also applied familiar copyright limits: the facts reported in an article are not protectable, and reproducing subject matter or headline-level information does not amount to competing with the original.
On training itself, the judge treated the ingestion of text for model training as falling within Indian copyright law's exception for private and research use, and assessed the activity as transformative under a three-part fairness test.
What remains open
The order is interim and leaves the harder questions for trial: whether retrieval-augmented outputs that surface current articles constitute communication to the public, and whether trained weights amount to a lasting store of the underlying works. Those issues, not training-set ingestion, are where news publishers' strongest arguments now sit.
Why it matters
India is the largest single market for ChatGPT by users, and ANI v. OpenAI has been the reference case for how a major non-Western jurisdiction outside the EU would treat AI training. A ruling that reads training as transformative and research-adjacent removes near-term legal risk from OpenAI's largest growth market and gives every developer operating there a citable precedent. It also demonstrates how thin the evidentiary base for these claims can be: the case turned less on doctrine than on a plaintiff unable to show the specific works in the specific model.