Can AI Learn from Your Content? ANI v OpenAI and India’s Proposed Licence for AI Training

Generative AI is trained on text, images and sound that other people created. Whether that needs their permission is one of the most consequential questions in copyright law today, and in 2026 India got its first answer from a court, at least for now. On 24 July 2026 the Delhi High Court refused the news agency ANI an interim injunction against OpenAI. Seven months earlier, a committee set up by the Department for Promotion of Industry and Internal Trade (DPIIT) had proposed a different answer: let AI developers train on everything that is lawfully available, but make them pay. This article explains both.

The dispute

ANI sued OpenAI in the Delhi High Court in 2024 (CS(COMM) 1028/2024). It made two claims. The first, which the court called the “training claim”, was that OpenAI copied and stored ANI’s news reports to train the large language models behind ChatGPT. The second, the “output claim”, was that ChatGPT’s answers reproduced ANI’s work. ANI asked for an interim injunction until the suit is tried. The court heard OpenAI, several interveners and two amici curiae, Professor Arul George Scaria and Mr Adarsh Ramanujan, reserved judgment on 27 March 2026, and dismissed the application on 24 July 2026 (Justice Amit Bansal).

What the court held

Jurisdiction. OpenAI argued that training happened on servers outside India, so Indian copyright law could not apply. Prima facie, the court did not accept that it lacked jurisdiction, or that the training claim would require Indian law to apply abroad (para 55).

The answers ChatGPT gives. The examples in ANI’s plaint were generated after the models had been trained, so they could not show that the models had memorised ANI’s reports, and the answers were not a substantial reproduction of ANI’s articles. Whether the models memorise and repeat training data is a disputed question of fact for the trial (paras 124 and 125).

Training is, prima facie, fair dealing. Section 52(1)(a)(i) of the Copyright Act, 1957 allows fair dealing with a work for “private or personal use, including research”. The court applied two tests.

• Purpose. The exception is not limited to non-commercial use: several other clauses of section 52 say “non-commercial” in terms, and this one does not. That a use is commercial does not by itself take it outside fair dealing (para 179).
• Fairness. The court asked whether the use was limited to training, whether it competed with ANI’s business or damaged it, and whether it served the public interest (para 236). ANI had placed nothing on record to show lost subscribers or revenue, and the court found considerable public benefit in trained language models, in education, research, accessibility and access to information (paras 250 to 254).

Both tests were met, so, prima facie, OpenAI’s storage of ANI’s works for training was fair dealing and not infringement (para 256).

Balance of convenience. Two facts weighed heavily. ANI could have blocked OpenAI’s web crawlers and had not; OpenAI said it had since blocked ANI’s website itself. And ANI had offered OpenAI a licence for USD 7.5 million in October 2024, which showed that any loss could be compensated in money. An injunction, by contrast, would harm OpenAI and the public, including the millions of people in India who use ChatGPT (paras 262 to 270).

What the ruling does not decide

These are prima facie findings on an interim application. The court said expressly that they have no bearing on the final outcome of the suit (para 274). ANI has appealed, and a Division Bench of the High Court issued notice on the appeal in September 2026. Until the appeal and the trial are decided, the ruling is a strong indication, not settled law.

The government’s proposal: one licence, one payment

In April 2025 DPIIT set up a committee to examine whether the Copyright Act is equipped for generative AI. Part 1 of its working paper, “One Nation One License One Payment”, was published for comments in December 2025. By a majority, the committee rejected both free use and a purely permission-based system, and recommended a hybrid model:

• a mandatory blanket licence allowing AI developers to train on all lawfully accessed copyright works, without an option for rights holders to withhold their works;
• a statutory right for rights holders to be paid a percentage of the revenue from AI systems trained on their works, at rates fixed by a committee appointed by the government and open to judicial review; and
• a single non-profit collecting body, the Copyright Royalties Collective for AI Training (CRCAT), whose members would be copyright societies and collective management organisations, paying rights holders who register their works whether or not they belong to a society.

Nasscom, a member of the committee, dissented: it favoured a text and data mining exception from which rights holders could opt out. A second part, on the output of AI systems, is to follow. Nothing has been enacted, and the proposal sits uneasily with the High Court’s view that training may already be fair dealing. Both are likely to shape the law over the coming year.

What to do now

• Publishers and creators. Decide whether you want your content used for AI training. If not, block AI crawlers (through robots.txt and similar controls) and keep a record of when you did so: the court gave weight to ANI’s not having done this. Keep evidence of where your content appears, and of any loss of subscriptions or licensing income.
• Licensing. Licensing on your own terms remains open. Bear in mind that an offer to license shows that money can compensate you, which counts against an interim injunction.
• AI developers. Keep records of the sources of training data and how they were accessed; respect paywalls, logins and opt-outs; and test outputs for verbatim reproduction. Each point mattered in this case, and each would matter under DPIIT’s model.
• Businesses using AI tools. Check what your provider’s terms say about intellectual property claims and indemnities, and review AI-generated material before you publish it.

See also our pages on copyright and technology, media and entertainment, and our article on consent, AI and personal data under the DPDP Act.

This article is for general information and states the law as at 2 November 2026. It is not legal advice. For advice on your situation, contact the firm. Published by BRB Legal.