Key takeaways
- A Delhi High Court single judge refused an interim injunction in ANI v. OpenAI in July 2026 and viewed training as arguably fair dealing, but the ruling is interim and on appeal.
- The DPIIT working paper proposes a mandatory blanket licence with statutory royalties and no opt-out; it is not law yet.
- Under Section 17 a newspaper proprietor is first owner only for publication in the newspaper unless an agreement says otherwise, so chain of title needs checking.
- Use arbitration and a change-in-law clause, because a US court judgment is not directly enforceable in India.
Indian publishers sit on an attractive pool of training material: news archives, books, magazines and regional-language content that large language models see far less of than English-language sources. US AI companies that want it are asking the same questions. Do we need a licence at all? Who can grant it? And what happens if India changes its law in the middle of the contract?
The Indian legal position on AI training is moving on several fronts at once: a court case with a July 2026 ruling and a pending appeal, a government working paper proposing a mandatory licence, and a Copyright Act that has no text-and-data-mining exception. This guide explains where things stand, why chain of title is the hardest part of an Indian publisher deal, what to put in the contract, and how to treat the risk of training without a licence. It’s written from the Indian-law perspective; US copyright and contract points belong to US counsel.
Why the legal position is unsettled
India’s Copyright Act, 1957 gives copyright owners exclusive rights to reproduce and adapt their works. It has no express exception for text and data mining, and the fair dealing provisions in Section 52 are tied to listed purposes such as private use, criticism, review and reporting. Whether copying material for AI training fits within those purposes is the question now before the courts.
The leading case is ANI v. OpenAI in the Delhi High Court, which ANI Media filed in November 2024. After more than eighteen months of hearings, a single judge dismissed ANI’s application for an interim injunction on 24 July 2026 and said that, at first sight, training an AI model on copyrighted data can be fair dealing. ANI has appealed, and on 15 September 2026 a Division Bench issued notice to OpenAI, declined to give ANI immediate relief and listed the appeal for 8 December 2026 (LiveLaw; Bar & Bench).
For a US AI company, the lesson is that the first Indian ruling leans in favour of training but is interim, is on appeal, and is limited to the facts of one case. It doesn’t tell you what a court would do with a different type of content or with the output side of the claim, such as a chatbot reproducing text. A licence buys certainty and removes the argument, and it remains valuable even if the single judge’s view prevails.
The policy backdrop: the DPIIT working paper
On 8 December 2025, a committee set up by the Department for Promotion of Industry and Internal Trade issued Part I of a working paper on generative AI and copyright. Its central proposal is a hybrid model built on a mandatory blanket licence: AI developers could train on any lawfully accessed copyrighted work without negotiating individual licences, and creators would receive a statutory right to remuneration at government-set rates (Takshashila; Storyboard18; Mondaq).
The proposal has several features that matter for a US company. Payments would go through a new central body, the Copyright Royalties Collective for AI Training (CRCAT). Developers would have to file disclosures about what they trained on. A rate-setting committee would fix the royalty. And copyright owners would have no right to opt out, which distinguishes the proposal from the European approach. The committee was not unanimous: Nasscom dissented and favoured a text-and-data-mining exception with an opt-out (Onmanorama; Takshashila).
None of this is law yet. A May 2026 commentary notes that no amendment has been introduced, and a second working paper on AI outputs is expected separately. The practical point for your contracts is that a licence you sign today may be overtaken by a statutory scheme in a year or two. Build for that, with a change-in-law clause that lets you adjust the commercial terms if a statutory licence or royalty scheme is introduced.
Who can actually license: chain of title
The commercial headline of an Indian publisher deal is the price. The legal issue that most often derails it is who actually owns the rights. Publishers frequently don’t own everything in their archives. Under Section 17 of the Copyright Act, an author is the first owner of copyright unless a specific rule says otherwise. For works made by an author employed by a newspaper or magazine under a contract of service, the proprietor is the first owner only in relation to publication in the newspaper or magazine, and the author is the first owner in all other respects, in the absence of a contrary agreement. So a newspaper may not own the right to license its journalists’ articles for AI training unless its employment contracts say so.
Assignments add another layer. Under Section 19, an assignment must be in writing, identify the work and the rights, and state the duration and territory. If duration isn’t stated, it is deemed to be five years, and if the territory isn’t stated, it is presumed to be India. An assignment can also lapse if the assignee doesn’t exercise the rights within a year, unless the agreement provides otherwise. Many older author agreements were drafted without AI in mind and with these defaults unexamined. A licence from the publisher that doesn’t match the underlying rights is a licence of something the publisher can’t give.
Add the other third-party content in an archive: photographs from agencies, syndicated material, freelance contributions, letters and user comments. Authors also have moral rights under Section 57. Before paying for a dataset, ask for a schedule of the rights the publisher actually holds, representations and warranties about ownership, an indemnity for third-party claims, and a right to exclude content that is shown to be unlicensed. Indian publishers can usually only warrant what they’ve checked, so budget for a diligence exercise.
Key contract terms
Define the licensed uses precisely. Pre-training, fine-tuning, retrieval-augmented generation and displaying excerpts in answers are different uses with different value and different legal risks, and a clause that says “for AI purposes” will be argued over. Say whether the licence covers the model’s outputs, whether the model may reproduce passages, whether attribution or links are required, and what happens to the licensed content after the term ends. A model can’t be made to forget, so the usual compromise is that the trained model may be kept and used, while the dataset must be deleted.
Term, price and reporting come next. Choose between a flat fee, a per-use payment or a revenue share, and expect Indian publishers to ask for reporting on how their content is used, which anticipates the disclosure the DPIIT proposal would require. Include a change-in-law clause, a most-favoured-licensee clause if you want one, and audit rights both ways. Be specific about exclusivity: a non-exclusive licence is normal, and an exclusive one is expensive and raises competition questions.
| Term | What to decide | Why it matters |
|---|---|---|
| Licensed uses | Pre-training, fine-tuning, retrieval and display of excerpts | Different uses carry different value and risk |
| Term and retention | Whether the trained model may be kept when the licence ends | A model cannot be made to forget |
| Payment and reporting | Flat fee, per-use or revenue share; usage reporting | Anticipates the disclosure the DPIIT proposal would require |
| Change in law | Right to adjust terms if a statutory licence or royalty scheme is introduced | Protects against being overtaken by regulation |
| Disputes | Arbitration with a neutral seat | A US judgment is not directly enforceable in India |
Plan for disputes. India does not treat the United States as a reciprocating territory for the direct execution of court judgments, so a US court judgment against an Indian publisher is not directly enforceable in India. An arbitration clause is generally the better route, because arbitral awards from a New York Convention country can be enforced in India. Choose a neutral seat and institution, and make sure the clause survives termination. For payments, the Indian publisher is the taxpayer in India on its receipts, and exports of services can raise their own goods and services tax and foreign exchange questions on the Indian side, which the publisher’s advisers should handle.
Risks, remedies and quick answers
If you’ve already trained on Indian content without a licence, don’t wait for a claim. Map what you used and where it came from, check whether any of it was behind a paywall or subject to a robots instruction, identify personal data in the dataset (Indian data protection rules can apply to data about people in India), and decide whether to approach the main publishers for a retrospective licence. The ANI ruling is helpful, but it’s an interim decision on appeal and is limited to its own facts. If you receive a claim, preserve records, avoid public statements about the merits and consider an early settlement that includes a forward licence. Courts have been willing to hear AI cases in detail, and the first indication of how they treat outputs will come from the later stages of the same litigation.
Related reading: ANI v. OpenAI and training on copyrighted data, the DPIIT AI copyright working paper, AI-generated content and copyright: India vs the US, who owns AI-generated content in India, Indian AI teams and US client data, who owns the AI model, weights and datasets and the EU AI Act and Indian SaaS companies; also copyright in legal headnotes and AI training and structuring AI training data licences after Ross.
Quick answers
Is AI training on Indian content legal without a licence?
It is unsettled. In ANI v. OpenAI, a single judge refused an interim injunction in July 2026 and viewed training as arguably fair dealing, but the ruling is interim and on appeal, and a government working paper proposes a mandatory paid licence.
Will a licence I sign now be overtaken by a statutory scheme?
Possibly. The DPIIT working paper proposes a blanket licence with statutory royalties. Include a change-in-law clause so that the commercial terms can be adjusted if India adopts it.
Who owns the rights to a newspaper article for AI training?
It depends on the contract. Under Section 17, a newspaper proprietor is first owner only for publication in the newspaper unless an agreement says otherwise, so check the author agreement.
Does India have a text-and-data-mining exception?
No. The Copyright Act, 1957 has no express exception for text and data mining, and the fair dealing provisions in Section 52 are tied to listed purposes. The DPIIT working paper proposes a mandatory licence instead of an exception.
Can I rely on a publisher’s warranty of ownership?
Only to a point. Publishers often do not own every right in their archives, so ask for a schedule of rights held, representations and warranties on ownership, an indemnity for third-party claims and a right to exclude content shown to be unlicensed.
Final word
Halverton & Co. is an Indian law firm advising US companies, founders, investors and law firms on Indian law, including copyright and AI licensing in India, chain-of-title diligence, technology contracts and disputes with Indian counterparties. We practise in Jharkhand, Maharashtra and before the Supreme Court of India, and work alongside US counsel, who advise on US law. Halverton & Co.: Where tech needs law! If you have a question about Indian law, write to us at office@halvertonandco.com, or get in touch.
This article reflects developments reported up to late October 2026. It is for general information only, is not legal advice, and does not create an attorney-client relationship. Halverton & Co. is an Indian law firm and does not advise on US law; US-law points should be confirmed with US counsel.
Sources
Related practice area
Technology & AI Law
Contracts, data protection and AI policies for products that handle other people’s data.
View the practice areaComments
No account needed. Comments are reviewed before they appear, so yours will show here once approved. Please do not share confidential details or describe your own legal matter: replies are general discussion, not legal advice, and do not create an advocate-client relationship. For your own situation, write to office@halvertonandco.com.
