On 24 July 2026, the Delhi High Court refused ANI Media’s request for an interim injunction against OpenAI, holding — for now — that training an AI model on copyrighted content is prima facie “fair dealing” under Section 52(1)(a) of the Copyright Act, 1957, and does not amount to infringement. Justice Amit Bansal’s 135-page order is India’s first substantive judicial treatment of AI training and copyright. It is an interim finding, not a final judgment — the suit continues — but it is already the most important signpost Indian law has offered on the question.
Here is what the Court actually held, what it deliberately left open, and what it means practically for content creators, publishers, and AI companies in India.
What was the dispute about?
ANI (Asian News International), one of India’s largest news agencies, sued OpenAI alleging that:
- Training infringement — ChatGPT was trained on ANI’s copyrighted news reporting without a licence;
- Output infringement — ChatGPT could reproduce portions of ANI’s reporting in its answers; and
- Reputational harm — ChatGPT sometimes generated fabricated statements wrongly attributed to ANI (“hallucinations”).
ANI sought an interim injunction restraining OpenAI from storing or using its works. OpenAI argued its use was transformative, that its infrastructure sits outside India, and that training does not reproduce works in the copyright sense.
What did the Court hold?
1. Training is prima facie fair dealing. The storage and use of copyrighted material to train a model was held, at this interim stage, to fall within Section 52(1)(a) — the exception for “private or personal use, including research”. On this reasoning, learning statistical patterns from lawfully accessed works is not the kind of “reproduction” copyright law prohibits.
2. No substantial reproduction was shown in outputs. The Court found ANI had not demonstrated that ChatGPT memorised and regurgitated its original literary works. Isolated or prompted snippets were not enough; the outputs did not amount to substantial reproduction of ANI’s articles.
3. Indian courts have jurisdiction over offshore AI companies. OpenAI’s servers being abroad did not defeat the claim: where a product is offered in India and its effects are felt in India, the Indian Copyright Act applies. Companies cannot locate infrastructure offshore and claim immunity.
4. It is interim, and key questions are open. The Court expressly left open the questions of memorisation, and of reproduction through retrieval-augmented generation (RAG) — where a model fetches and summarises live content rather than relying on training. The final trial will decide these.
Why this ruling matters
- It is India’s first analytical framework for AI and Section 52. Until now, Indian creators and AI companies were reasoning from US “fair use” cases. India’s fair dealing exception is narrower and enumerated — and the Court has now shown how it can stretch to AI training.
- It aligns India, for now, with the global direction. US courts have recently leaned toward treating training as transformative use. Publishers worldwide are losing the training battle and shifting to licensing negotiations and output-side claims.
- It arrives while Parliament reconsiders the statute. The Commerce Ministry has constituted an expert panel to review the Copyright Act’s fitness for AI. The final word may come from the legislature, not the courts.
What it means for content creators and publishers
- Your registered copyright still matters — arguably more. The Court’s reasoning turned on what ANI could prove: no demonstrated memorisation, no substantial reproduction. Creators who can evidence ownership and copying stand in a far stronger position. Registration gives you prima facie proof of ownership and a clean chain of title — the foundation for any output-side claim. See our copyright registration services.
- Output claims remain fully alive. If an AI tool reproduces your work substantially — your article, your song, your code — that is still actionable. The ruling protects training, not regurgitation.
- Hallucinated attribution is a separate, live harm. False statements attributed to you may sound in defamation and personality rights even where copyright fails. The Bombay High Court’s Arijit Singh line of cases shows Indian courts will protect persona and attribution vigorously.
- Licensing is becoming the real market. As training-side claims weaken, the commercial answer is negotiated licences for high-value content. Publishers with organised, registered catalogues will negotiate from strength.
What it means for AI startups and businesses using AI
- Training on lawfully accessed data has meaningful cover — for now. An interim, appealable order is not a safe-harbour statute. Document your data provenance; scraping behind paywalls or in breach of terms remains risky.
- RAG pipelines are expressly an open question. If your product fetches and summarises third-party content live, ANI does not protect you. Build attribution and licensing into retrieval features.
- Offshore structuring will not shield you in India. If your AI product serves Indian users, expect Indian copyright jurisdiction.
- Watch the appeal and the expert panel. Both could move the line within a year. Contracts, indemnities and data-sourcing policies should anticipate change.
The bigger picture
ANI v. OpenAI is one interim order in a case that will likely reach the Division Bench and perhaps the Supreme Court, in parallel with legislative review. The near-term equilibrium it creates, though, is clear: training is presumptively permitted; outputs and attribution are where the fights will be. For India’s creator economy — the YouTubers, writers, musicians, and software teams we work with — the practical takeaway is not that copyright lost, but that provable, registered, well-documented copyright is the asset that survives the AI era.
Frequently asked questions
Did OpenAI win the ANI case?
OpenAI won the interim round: the Delhi High Court refused ANI’s injunction, holding that training is prima facie fair dealing. The full suit continues, and the findings can change at trial or on appeal.
Does this mean AI companies can freely use my content in India?
For training, the interim answer leans yes where content was lawfully accessed. But substantial reproduction in outputs, live retrieval (RAG) of your content, and false attribution remain actionable.
Is “fair dealing” the same as US “fair use”?
No. India’s Section 52 lists specific permitted purposes; US fair use is an open-ended four-factor test. This ruling stretches “private or personal use, including research” to cover model training — a notably broad reading that the appeal may test.
What should I do to protect my content from AI misuse?
Register your copyrights (prima facie proof of ownership), watermark and document your publication trail, monitor for substantial reproduction in AI outputs, and take advice before licensing. If an AI tool reproduces your work or fabricates statements in your name, a cease-and-desist and, where needed, court action remain available. See copyright registration and copyright objection support, or talk to an attorney.
Can Indian courts hear cases against foreign AI companies?
Yes — the Court confirmed jurisdiction where the product is offered and its effects are felt in India, regardless of where servers sit.
This article is for general information and education only and is not legal advice. For advice on your specific situation, consult a qualified IP attorney.
The InPASS Patent E-register tab: enter the granted patent number and CAPTCHA
The InPASS Application Status tab: enter the application number and CAPTCHA