The Delhi High Court has given OpenAI an important interim win in a copyright fight brought by Indian news agency Asian News International (ANI). Judge Amit Bansal rejected ANI's request for a preliminary injunction, finding that the agency had not proven the core claims needed for immediate relief.
The ruling does not end the dispute. The court will still examine unresolved questions in the main proceedings, including how Retrieval Augmented Generation (RAG) should be treated and whether outputs based on real-time retrieval can qualify as "communication to the public." But at this stage, ANI failed to show that OpenAI's systems copied its journalism in the way the agency alleged.
Why ANI's evidence failed
ANI argued that ChatGPT produced substantial copies of its articles. OpenAI responded with a timing problem that weakened the claim: the models at issue, GPT-4 and GPT-4o, were trained on data from April 2022 and April 2024, while the articles ANI relied on were mostly from August and September 2024.
That meant those cited articles could not have been included in the training data for those models. The judge's preliminary view was that any similarities were more likely connected to RAG, which allows a model to retrieve online information in real time, much like a search engine.
ANI had not addressed RAG in its filing, so the court did not make a final ruling on that question. The judge did, however, leave open the possibility that RAG-based outputs may raise separate legal issues later in the case.
The court also looked closely at how ANI tested ChatGPT. The agency used adversarial prompts, telling the model to reproduce articles "exactly." Even with that instruction, ANI did not produce a single verbatim copy.
Facts, headlines and competition
The ruling drew a line between protected expression and unprotected facts. The judge found that facts in news articles in general are not copyrightable. On the evidence before the court, reproducing topics and headlines did not amount to direct competition with ANI.
That mattered because ANI needed to show more than similarity. It needed to show that OpenAI's conduct caused the kind of harm that would justify a preliminary injunction. The court found no economic harm because OpenAI and ANI operate in different sectors.
The judge also rejected ANI's argument, for now, that OpenAI permanently stores training data inside its models and can reproduce ANI's work verbatim on demand. The evidence did not support that claim at the interim stage, though the court will revisit the issue in the main proceedings.
How the court viewed AI training
ANI also challenged the use of its work in AI training. Both parties agreed that OpenAI had used ANI content during training, but OpenAI argued that the material was only a tiny share of the overall dataset and that the model extracted non-expressive elements such as grammar, syntax, and language patterns.
The judge considered exceptions under Indian copyright law and relied on a clause covering "private or personal use, including research." He read "research" broadly enough to cover AI training, at least at this stage of the case.
That reading came with limits. Training copies must come from lawful sources, not shadow libraries or paywalled sites accessed without permission. OpenAI also did not make the training copies public and processed them only internally.
AI copyright law expert Andres Guadamuz called the ruling an important early win for OpenAI and said it is the first time a court has explicitly found that AI training falls under a private use exception.
The fairness test favored OpenAI
The court applied a three-part fairness test and sided with OpenAI on all three counts. OpenAI's use of ANI's work was limited to training because ANI had not proven memorization or reproduction. The court also found no demonstrated economic harm because the companies operate in different sectors.
Even when users ask ChatGPT about ANI headlines, the model returns topics and, at most, a few article titles. That did not persuade the court that ChatGPT was replacing ANI's business in this case.
The judge cited U.S. cases including Bartz v. Anthropic and Kadrey v. Meta, where language model outputs were treated as transformative. He also pointed to the earlier Google Books ruling. In the court's view, trained language models can improve access to information, support education, advance scientific research, help with software development, enable translation, and create tools for people with disabilities.
What remains unresolved
The Delhi ruling joins a global set of AI copyright cases that point in different directions. In the U.S., Raw Story and AlterNet's lawsuit against OpenAI was thrown out because the plaintiffs could not show enough harm and the odds of exact copies were low. The GitHub Copilot case also failed because plaintiffs could not present a single example of identical code.
Other cases have been less favorable to AI companies. The Intercept won a partial victory through a DMCA complaint over copyrighted material that had been stripped out before training. In Ross Intelligence v. Thomson Reuters, a court denied fair use because the AI research tool directly competed with Thomson Reuters' legal database Westlaw, though the court said that ruling applied only to that non-generative use case and could not be extended to large language models.
In Europe, the Munich Regional Court ruled in the GEMA case that song lyrics were reproducible in model weights, while the High Court in London dismissed the Getty Images v. Stability AI lawsuit and ruled that an AI model is not an "infringing copy." The result is a legal landscape where AI training, model memorization, lawful data sources and adversarial prompts remain central disputes.
For news agencies, the business concern reaches beyond copyright. A Pew Research Center study cited in the source article found that click-through rates to external websites drop to just 8 percent with Google's AI Overviews, compared to 15 percent without an AI summary. That means AI systems may still pressure the news business even where courts do not find direct copyright infringement.