The same week CNN became the latest newsroom to drag an AI company into court, India's own answer to the question CNN is asking — who pays when a model ingests journalism? — moved closer to becoming law. The two developments are not coincidental. They are halves of a single global negotiation over the price of training data, and India is positioned to resolve it by statute before American courts resolve it by litigation.
A widening American front
On May 28, 2026, CNN sued Perplexity AI in the US District Court for the Southern District of New York, alleging the company scraped and copied more than 17,000 CNN stories, photos and videos and fed them, in real time, into the large language model behind its answer engine. CNN says the two sides discussed a licensing deal in 2024, could not agree on terms, and that CNN then blocked Perplexity's crawler — after which the copying allegedly continued. A second count targets Perplexity's "Comet Plus" tier, which CNN claims falsely implied an ongoing partnership. Perplexity's reply was blunt: "You can't copyright facts."
It is CNN's first AI copyright action and reportedly the first by a television network, joining The New York Times, News Corp and the New York Post. That "facts" defence is the crux of every one of these cases: copyright protects expression, not the underlying information, and a model that learns statistical patterns is arguably doing something closer to reading than to reproducing.
India's parallel reckoning
That exact argument is already before the Delhi High Court. In November 2024, news agency ANI sued OpenAI, alleging ChatGPT was trained on its copyrighted reporting without permission; OpenAI says it blocklisted ANI's domain in October 2024. The Federation of Indian Publishers intervened on January 8, 2025, and the Digital News Publishers Association — representing much of India's mainstream press — on January 28. The court framed four questions: whether storing data to train infringes, whether generating answers infringes, whether Section 52 fair-dealing applies, and whether an Indian court even has jurisdiction over a US firm. Arguments have closed and judgment is reserved.
Section 52 is the structural reason the case matters. Unlike the open-ended US fair-use standard, Section 52 of the Copyright Act, 1957 is a closed, enumerated list — private study, criticism, review, news reporting — and contains no text-and-data-mining (TDM) exception of the kind the EU and Japan have enacted. A purpose-specific list is far harder to stretch around model training than America's four-factor balancing test, which means an ANI win could render large-scale training in India presumptively unlawful in a way it is not in the US.
DPIIT's hybrid gamble
New Delhi has not waited for the court. On December 8, 2025, the Department for Promotion of Industry and Internal Trade published a Working Paper on Generative AI and Copyright proposing a "hybrid" licensing model. Under it, AI developers would automatically receive the right to train on copyrighted works; rights holders could not opt out; and royalties — triggered only at commercialisation, not at training — would be collected and distributed by a new statutory body, the Copyright Royalties Collective for AI Training (CRCAT), which would maintain a registry developers query before ingesting data.
The case for this design is serious and deserves to be stated plainly. A pure opt-out or market-licensing regime advantages incumbents who can afford to paper the world with deals — exactly the dynamic that let OpenAI and Google sign with major publishers while smaller outlets got nothing. A compulsory licence with centralised collection guarantees that even a freelance photographer or a regional Hindi daily gets paid, and it spares startups the impossible task of negotiating millions of bilateral licences. That is a real equity argument, and the panel is right that the news industry's survival concern is not imaginary.
But the cure risks being disproportionate to a disease still being diagnosed. Stripping rights holders of any ability to say "no" — converting copyright into a compulsory royalty stream by fiat — is a heavier intervention than even the EU made; Brussels built its TDM exception around an opt-out, not its abolition. India's royalty collectives have a poor record: the existing music body has spent years mired in disputes over opaque distribution. Bolting AI training onto that model invites the same opacity at far greater scale. And mandating that every developer ping a government registry before training is a compliance tax that lands hardest on the very Indian startups DPIIT says it wants to protect — while the largest foreign labs absorb it easily.
A proportionate path
The better sequence is to let the Delhi High Court rule first, then legislate to the gap it actually exposes — not a hypothetical one. If the court reads Section 52 narrowly, Parliament should weigh a genuine TDM exception with a meaningful opt-out, as the EU and Japan have, rather than a no-consent compulsory licence run by a new collective. That preserves the bargain CNN is fighting for in New York: publishers who want to license, can; those who want to block, can — without erecting a permanent royalty bureaucracy on the assumption that markets cannot function. India's AI ambitions and its creators are not on opposite sides. A regime that pays creators while keeping consent intact serves both; one that abolishes consent to guarantee payment trades away more than it secures.