The Delhi High Court resumed hearings this month in the consolidated copyright suits filed against OpenAI by Asian News International (ANI), the Federation of Indian Publishers — including Bloomsbury India, Penguin Random House India, and Cambridge University Press — and Indian music labels led by T-Series and Saregama. The proceedings, the most consequential AI litigation anywhere in the Global South, will determine two questions with far-reaching consequences for India's AI ambitions: whether training a large language model on copyrighted Indian works without a licence is infringement under the Copyright Act, 1957, and whether an Indian court has jurisdiction over a US company whose servers sit outside India.
How the bench answers will shape whether India builds frontier AI at home or licenses it, line by line, from foreign labs.
The case in brief
ANI filed first, in November 2024, alleging that ChatGPT had been trained on its newswire content and was reproducing it on demand. The Federation of Indian Publishers and the music industry plaintiffs followed through 2025, broadening the dispute from journalism into books and sound recordings. OpenAI's primary defences are well-rehearsed in similar US and EU cases: training is a transformative, non-expressive use; outputs are statistical generations, not copies; and in any event, an Indian court lacks jurisdiction because no training, hosting, or model-serving infrastructure is located in India.
The jurisdictional question alone is novel. OpenAI's position — that mere accessibility of ChatGPT from India does not establish a forum — runs into a line of Indian rulings stretching back to (India TV) Independent News Service v. India Broadcast Live and the Supreme Court's reasoning in World Wrestling Entertainment v. Reshma Collection, which have treated commercial targeting of Indian users as sufficient. If the Delhi High Court adopts that framing, every foreign AI lab serving Indian users becomes amenable to Indian courts. That is a significant assertion of long-arm jurisdiction, and not one OpenAI alone should fear.
The Copyright Act was not written for transformers
The substantive question is harder. India's Copyright Act, 1957 contains a closed list of permitted uses in Section 52 — not an open-textured fair use doctrine of the American kind. The closest analogues are fair dealing for research, review, and reporting of current events. None of these clauses contemplates the ingestion of billions of works into a statistical model.
That gap is not a reason to declare training categorically unlawful. It is a reason to update the statute. The European Union, often caricatured as hostile to AI, in fact codified a text-and-data-mining (TDM) exception in Articles 3 and 4 of the 2019 Copyright in the Digital Single Market Directive, allowing TDM on lawfully accessed works subject to a rightsholder opt-out. Japan's Article 30-4 of its Copyright Act goes further, permitting machine learning on copyrighted works where the use is not for the enjoyment of the expression itself. Singapore added a computational data analysis exception in 2021. Each regime balances innovation with rightsholder protection through proportionate carve-outs — not blanket prohibition.
What a maximalist ruling would cost India
If Delhi rules that ingestion equals infringement and that every model serving Indian users must license every work in its corpus, the practical consequences are stark.
- Domestic models suffer most. Indian startups building on Indic-language corpora — Sarvam, Krutrim, and a wave of vernacular LLMs — depend on local training data far more than OpenAI does. They have the least leverage to negotiate collective licences.
- Linguistic exclusion. India has 22 scheduled languages. A licensing-only regime privileges English-language and well-organised rightsholders, leaving Tamil, Bengali, Marathi, and tribal languages systematically under-represented in training data because they lack the collective licensing infrastructure to bargain.
- Capital flight. As Rest of World reported on 15 May 2026, Indian VCs are now outpacing Silicon Valley at home, with local funds leading the majority of early-stage rounds. A hostile copyright climate would push the model layer of that capital back overseas just as it is finally indigenising.
A proportionate path forward
The court need not — and should not — choose between rightsholders and innovators. Three principles can guide a measured ruling:
First, distinguish ingestion from reproduction. The harm rightsholders complain of is not training per se; it is verbatim regurgitation in outputs. Output-side liability is well within existing copyright doctrine. Input-side liability is not, and importing it would freeze a technology that is already serving Indian users.
Second, recognise jurisdiction without weaponising it. Indian courts can and should assert authority over foreign platforms serving Indian users — but the substantive law applied must be capable of accommodating modern technology, not punishing it.
Third, push Parliament to act. The Copyright Act has been amended thirty-eight times since 1957. A TDM exception modelled on the EU's Article 3-4 structure, with a clear opt-out mechanism, would resolve the dispute prospectively and give Indian developers the legal certainty European and Japanese rivals enjoy.
The plaintiffs' grievances are real. So is the cost of getting this wrong. India is the rare jurisdiction where domestic capital, domestic talent, and domestic data could combine to produce frontier AI on Indian terms. Delhi has a chance to write a copyright ruling worthy of that opportunity — proportionate, jurisdictionally confident, and innovation-friendly. The alternative is a future in which Indians use AI built somewhere else, on someone else's data, under someone else's rules.