OpenAI wins early ruling in ANI copyright case in Delhi

The OpenAI ANI copyright case produced an early win for OpenAI after the Delhi High Court refused to grant Asian News International a preliminary injunction over ChatGPT training and outputs. The interim order matters for AI companies and publishers because Judge Amit Bansal signaled, at this stage, that ANI had not shown verbatim copying, market harm or unlawful public reuse by OpenAI.

ANI, one of India’s largest news agencies, sued OpenAI alleging that its copyrighted news content was used to train models and that ChatGPT could reproduce protected articles. The court denied immediate relief on both theories, while leaving some questions for the main proceedings.

What did the Delhi High Court decide in the OpenAI ANI copyright case?

The court found that ANI had not made the showing needed for an injunction. According to the judgment, ANI submitted ChatGPT outputs that it said copied its articles, but OpenAI showed that the relevant GPT-4 and GPT-4o training datasets closed in April 2022 and April 2024, while many of ANI’s cited articles were from August and September 2024.

That timing weakened ANI’s claim that the articles had been memorized from training data. The judge’s preliminary view was that any similarities may have come from retrieval augmented generation, or RAG, a method that lets a model pull online information at response time. ANI had not addressed RAG in its filings, so the court did not make a final ruling on that issue.

The court said RAG-based outputs may still raise a separate question over “communication to the public,” which will be considered later. That leaves a live issue for AI search and chatbot products that retrieve and summarize current web content, rather than relying only on model weights.

ANI’s evidence also included adversarial prompts that asked ChatGPT to reproduce articles “exactly.” Even under those conditions, the court said ANI had not produced a single verbatim copy. Judge Bansal also found, at this stage, that facts in news reports are not copyrightable and that producing topics or headlines did not amount to direct competition with ANI.

How the court treated AI training

Both sides accepted that ANI material had been used in training. OpenAI argued that the content represented a small part of the total training corpus and that the models learned non-expressive elements such as grammar, syntax and language patterns.

The judge relied on an exception in Indian copyright law covering “private or personal use, including research,” reading research broadly enough to cover AI training in this interim analysis. The court placed limits around that view: copies used for training must come from lawful sources, rather than shadow libraries or paywalled sites accessed without permission, and OpenAI had processed the material internally without making training copies public.

AI copyright law scholar Andres Guadamuz described the ruling as an early win for OpenAI and said it appears to be the first ruling to explicitly place AI training within a private use exception.

The court also found no economic harm on the current record because OpenAI and ANI operate in different sectors. The judge credited public-benefit arguments around language models, including education, research, software development, translation, accessibility and wider access to information.

Why publishers will keep fighting AI summaries

The Delhi ruling fits into a split global pattern. U.S. cases involving Raw Story and AlterNet against OpenAI, and a GitHub Copilot dispute, failed where plaintiffs could not show concrete copying. Other matters, including Ross Intelligence v. Thomson Reuters and litigation over pirated books used by Anthropic, have cut against AI defendants in narrower circumstances.

For publishers, the harder business problem may sit beyond training legality. The Decoder cited Pew Research Center findings that external click-through rates fall to 8% when Google AI Overviews appear, compared with 15% without an AI summary. If AI systems answer news queries well enough that users do not visit the original publisher, courts may face sharper arguments over competition and market harm in later cases.

This story draws on original reporting from The Decoder.