The Delhi High Court issued a ruling on 24 July 2026 in ANI Media v OpenAI that marks India's first significant judicial pronouncement on the legality of using copyrighted material to train large language models. Though the main case remains pending, the court's refusal to grant interim relief already indicates how Indian courts may approach AI training practices.
Key facts
- ANI Media alleges OpenAI used its news articles without permission to train ChatGPT
- The court rejected OpenAI's claim that Indian courts lack jurisdiction because servers are located in the US
- The court found Indian jurisdiction valid because ANI is based in India and outputs are generated for Indian users
- The case CS(COMM)1028/2024 is being heard with support from two amici curiae and six interveners
What the lawsuit centers on
ANI Media claims OpenAI used its news reports to train ChatGPT without authorization and that the model subsequently reproduces ANI's content. OpenAI argues that training constitutes lawful "fair dealing" under India's Copyright Act of 1957.
The court first explained in detail how large language models work: they predict sentence formation based on probability. Training data is tokenized (broken into words, phrases, and punctuation) and processed so the model learns statistical patterns. When responding to queries, the AI uses these patterns to generate responses—sometimes via Retrieval-Augmented Generation (RAG), where external information is retrieved during inference.
Jurisdiction: A precedent for global AI regulation
The ruling's core strength lies in rejecting OpenAI's jurisdictional defense. OpenAI had argued that US server locations and San Francisco jurisdiction clauses in its terms of service precluded Indian court authority. The court disagreed:
"In the present age of cloud computing and diminishing national boundaries in data storage, conventional concepts of territoriality cannot be strictly applied" – Delhi High Court, citing Neetu Singh & Anr v Telegram FZ & Ors (2022)
The judges characterized US servers as merely a "terminal link" in an inseparable chain: data accessed from India, processed, and outputs generated for Indian users. The rejection of strict territorial concepts was deliberate—otherwise companies could "evade Indian copyright law by shifting the terminal link to servers abroad."
What this means for German companies
This decision carries signal value far beyond India. It demonstrates that major markets will not accept that AI companies can hide behind infrastructure arguments. For German firms developing or training AI systems, the implication is clear: even if your servers are in the US or cloud-hosted, you may face liability in any jurisdiction where your models serve users or process data. The question of whether training on copyrighted material qualifies as fair use remains open—but jurisdiction is now settled.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
