A reversal worth reading slowly
On May 20, 2026, Justice Ravi Krishan Kapur of the Calcutta High Court dismissed IndiaMART InterMesh Ltd's interim relief application against OpenAI. Five months earlier, in an order dated December 26, 2025, the same judge had found a 'strong prima facie case' that IndiaMART was being 'selectively discriminated against' by ChatGPT, which surfaced links to competing B2B marketplaces but not IndiaMART's own. The December finding made headlines: an Indian court appeared willing to entertain a legal duty on a foreign AI lab to include a domestic platform in its outputs. The May order undoes that.
The case is the first Indian ruling on whether a domestic business has legal recourse against silent exclusion by a foreign generative-AI system. Its reasoning reaches well beyond the dispute between IndiaMART and OpenAI.
The steelman for compelled visibility
There is a serious argument behind IndiaMART's claim and it deserves an honest hearing. ChatGPT now has roughly 100 million weekly active users in India, second only to the United States, per Sam Altman's February 15, 2026 disclosure ahead of the India AI Impact Summit. Where 100 million people start their B2B supplier search increasingly shapes which platforms get demand. If OpenAI applies an opaque inclusion rule — IndiaMART says OpenAI relied on the 2024 USTR Notorious Markets List, which carries no Indian statutory force — and competing marketplaces named in the same USTR document still appear in answers, then a private decision in San Francisco has differential market effects in Indore.
That is a real problem if you take the view that scale converts editorial judgment into market gatekeeping. Once a conversational assistant becomes the dominant discovery interface for products and suppliers, the argument runs, exclusion from its outputs can materially affect market access. The question, in that telling, stops being one of speech and becomes one of platform neutrality.
Why the court got it right anyway
The Calcutta High Court engaged that argument and then declined it on three grounds, each faithful to existing Indian law.
First, on visibility as a right: 'No third party can compel a service provider to use its service in a manner to reflect its link or for its benefit,' the court said, citing the Delhi High Court's ruling in Google LLC v. DRS Logistics. Visibility on someone else's platform is not a justiciable entitlement absent statute or contract.
Second, on trademark dilution: Section 29(4) of the Trade Marks Act, 1999 requires use 'in the course of trade.' Silence — failing to surface a mark in an output — is not commercial use. The court declined to invent a tort of non-mention, observing that 'silence per se cannot constitute a cause of action.'
Third, and most consequentially for AI policy, on classification under the Information Technology Act, 2000. 'Originator,' defined in Section 2(1)(za), means a person who 'sends, generates, stores or transmits any electronic message… but does not include an intermediary.' Justice Kapur held that ChatGPT 'has an element of newness, uniqueness and originality in its results which ought to bring it within the definition of an originator rather than an intermediary.' The output is generated, not relayed.
What the originator label actually means
This characterisation cuts in two directions, and both deserve attention.
For OpenAI, it forecloses a regulator-style obligation of neutral carriage. India's intermediary regime — Section 79 safe harbour plus the 2021 IT Rules — was designed around platforms that passively transmit user content, where some duties of fairness are at least plausible. An originator generates its own speech. Compelling an originator to include a particular party's links is, in substance, compelled speech, and Indian courts have rightly resisted that move whether the compeller is the state or a private litigant.
The same label, however, means losing Section 79 safe harbour for downstream liability — defamation, copyright, IT Rules takedown obligations — when something in a model output is unlawful. The court was careful to note that the originator question is 'complicated and vexed' and will need expert evidence at final hearing. But on a prima facie view, the regulatory bargain is now visible: generative-AI providers will not be conscripted as common carriers, and in exchange they own what their systems say.
That is the correct division. Search engines that indexed third-party content sat naturally in the intermediary box. A system that composes a paragraph from learned patterns does not. Pretending otherwise — for the convenience of either regulators who want inclusion duties or providers who want safe harbour for hallucinations — would produce a regime that satisfies no one.
The wider signal to Indian AI policy
The Ministry of Electronics and IT and Parliament are circling a Digital India Act to replace the IT Act, 2000, and intermediary classification is one of its open questions. The Calcutta High Court has now planted a marker: the courts will not stretch 2000-era definitions to deem a large language model an intermediary simply because users treat it as a search box. Any duty of non-discrimination on generative platforms will have to come from a legislature, after public consultation, with a clear theory of harm — not from a writ asking a judge to order specific inclusions.
That is the proportionate answer. The market consequences of AI-mediated discovery are worth studying. If a remedy is warranted, it belongs in competition law applied through statute, in transparency obligations on training and inclusion rules, or in self-regulation under consumer-protection norms — not in writs asking judges to engineer specific model outputs.