India is working out whether AI companies can train on copyrighted work, and whether they must pay for it, from two directions at once. In court, the news agency ANI has been suing OpenAI in the Delhi High Court since November 2024. In policy, a committee under DPIIT, the commerce ministry's industrial-policy department, proposed in December 2025 that AI developers get an automatic licence to train on copyrighted works in exchange for mandatory royalties.
These point in different directions, and neither is final. This post sets out what each has actually decided or proposed, based on the court's order and the working paper as reported, and what it means for businesses using AI tools.
ANI v OpenAI: what the court decided in July 2026
ANI sued OpenAI in November 2024, alleging that ChatGPT was trained on its news content without permission and that it sometimes attributed made-up statements to ANI. On 24 July 2026, Justice Amit Bansal of the Delhi High Court ruled on ANI's request for an interim injunction — an order to stop OpenAI while the case continues — and refused it.
On jurisdiction, ANI won: the court held that Indian courts can hear a case against a global AI company where the effects are felt in India, even if training happened on servers abroad. On training, OpenAI won for now: the court found, on a prima facie basis, that storing and using ANI's content to train a model fell within the research exception in Section 52(1)(a)(i) of the Copyright Act, 1957, and could be fair dealing. On outputs, the court found ANI hadn't shown that ChatGPT reproduced its articles verbatim or substantially. ANI's complaint about false attribution — ChatGPT putting invented quotes in ANI's name — wasn't decided and remains live for trial.
The most important caveat is that this is an interim order. It refuses an injunction on a prima facie view; it does not finally decide that training AI on copyrighted news is lawful in India. The suit continues to a full trial, with two court-appointed amici — IP advocate Adarsh Ramanujan and Professor Arul George Scaria of NLSIU — assisting the court.
The government's plan: "One Nation One Licence One Payment"
On 8 December 2025, a DPIIT committee published Part I of a working paper on generative AI and copyright, titled "One Nation One Licence One Payment: Balancing AI Innovation and Copyright", and invited public comments within 30 days.
Its central proposal is a mandatory blanket licence: AI developers could use any lawfully accessed copyrighted work for training without asking permission, and copyright owners would get a statutory right to be paid. Owners could not opt out. Payments would be collected by a single government-designated non-profit — the Copyright Remuneration Collective for AI Training (CRCAT), made up of existing copyright societies and collective management organisations — and distributed to rightsholders whose works are registered for the purpose. Rates would be set through an institutional rate-setting mechanism with judicial review.
The committee said it had considered and set aside the alternatives on their own: a broad text-and-data-mining exception, an opt-out regime like the EU's, and purely voluntary licensing. Part II, on whether AI-generated outputs can be copyrighted and who the author would be, was expected in early 2026; we haven't found a published version as of this post. None of this is law yet — it would need amendments to the Copyright Act.
Why the two tracks point different ways
The court's interim view treats training as potentially fair dealing, which would mean no permission and no payment. The DPIIT proposal assumes training needs a legal basis and creates one — permission by statute, but with payment. If the court's reasoning holds at trial, AI companies will argue a licence regime is unnecessary; if the licence regime becomes law, it would largely settle the training question regardless of how the ANI case ends. Expect both to be argued over through 2027.
What this means if you use AI tools
For most businesses, the training dispute is between AI companies and rightsholders. The practical risks are on the output side.
Check what you publish. Neither the court ruling nor the proposal protects you if a tool's output copies someone else's work closely. For images, music and marketing copy, run a quick similarity check on anything prominent, and avoid prompting for a named living artist's style for commercial work.
Don't assume you own AI output outright. Whether purely AI-generated work is protected by copyright in India is exactly what Part II is meant to address. Until then, keep human creative input — selection, editing, arrangement — in anything you need to own, and keep records of it.
Read the indemnity. Several major vendors now offer IP indemnities on paid business plans, promising to defend customers against copyright claims over output. Adobe Firefly's commercial-safety positioning is one example in our catalogue. Check whether the plan you're on actually includes it; free and consumer tiers usually don't.
Watch for attribution errors. The part of the ANI case still alive is about AI inventing quotes and attributing them to a real organisation. If you publish AI-assisted content that quotes anyone, verify the quote at source.
What to watch
The ANI trial, and any appeal of the interim order; the final DPIIT recommendations and whether the government moves to amend the Copyright Act; and Part II on AI-generated works. We'll update this post as each moves. It's a plain-language summary, not legal advice — for decisions about specific content, talk to an IP lawyer.