The Calcutta high court’s recent decision in Indiamart Inter Mesh Ltd vs OpenAI Inc may appear at first glance to be a routine intellectual property dispute. An e-commerce platform alleged that ChatGPT’s responses were bypassing its listings and sought relief on grounds of dilution, disparagement and unfair trade practices. The court declined interim relief.
But the real significance of the case lies elsewhere. In assessing how ChatGPT functions, the court engaged with a larger question: can India’s intermediary liability framework, designed for a different era, meaningfully be applied to Generative AI? The answer, on the face of it, seems uncertain.
From passive platforms to generative systems: India’s intermediary framework under the Information Technology (IT) Act of 2000 was built on a stable understanding of how the internet works. Platforms were classified based on functions such as hosting content, transmitting data or facilitating transactions.
The safe-harbour provision of Section 79 protected such entities so long as they remained conduits of third-party information. The system rested on a simple assumption: the platform does not create any content; it enables access to it.
Generative AI disrupts that assumption. Large language models (LLMs) ingest vast datasets, apply probabilistic models and generate synthesized responses to assorted user queries. The output is not a list of links, but an answer, often authoritative in tone and sometimes entirely novel. This distinction is legally consequential. Once an online platform begins generating content rather than merely transmitting it, the foundation of intermediary law changes.
It is this shift that the Calcutta high court recognized, noting that systems like ChatGPT do not merely host or display content, but actively “generate,” “curate” and synthesize responses, functioning as active participants in the creation of information rather than passive conduits.
This led to an important observation in the judgement: “The question of whether ChatGPT falls within the definition of an ‘intermediary’ under section 2(1)(w) or an ‘originator’ under section 2(za) of the IT Act is a complicated and vexed question of both law and fact.”
The court also said that “in the dynamic nature of functions which ChatGPT performs, it curates, promotes and generates content. This makes it an active participant rather than a mere intermediary.”
These observations go to the heart of legal liability: intermediaries receive conditional safe harbour while originators do not. That said, these are passing observations, not a final determination.
India’s intermediary dilemma: The dilemma of AI regulation in India is structural. If LLMs are treated as intermediaries, they benefit from a safe harbour regime never designed for such systems.
At the same time, treating them as non-intermediaries creates an equally difficult problem. Generative AI systems are inherently probabilistic. They produce outputs dynamically and cannot feasibly pre-screen every response. If they fall entirely outside the intermediary framework, a difficult question of attribution arises.
The high court recognized this tension, observing that “any generation of an electronic record, service or response is always dependent on a user prompt” and that “the user supplies the intent and the platform executes it.” On this reasoning, liability may rest on the user rather than the AI system. Conversely, treating such systems as intermediaries would allow them to claim safe harbour immunity. Either classification would produce an untenable result.
A framework misaligned with technology: The basic issue is that India’s intermediary framework was never designed for this category of systems. Intermediary law evolved around message boards, social media platforms, search engines and e-commerce marketplaces, all operating on a common principle: they relate to third-party content. Generative AI breaks that model. Its outputs are neither entirely original nor entirely derivative. They are synthesized, contextual and often non-attributable. The intermediary-or-originator binary struggles to capture this reality.
The court acknowledged this limitation, noting that the IT Act predates modern Generative AI systems and that these questions may require legislative intervention.
This sits uneasily with India’s recent regulatory direction. The 2026 amendments to the Intermediary Rules bring AI-generated content within intermediary due diligence obligations, including labelling, disclosures and compliance for AI-enabled tools.
Notably, the primary impetus was the proliferation of AI deepfakes rather than a considered response to the broader question of how Generative AI systems ought to be classified. This suggests a regulatory approach that implicitly treats certain AI systems as intermediaries. Yet, the court’s observation on attribution points in the opposite direction.
Beyond the intermediary framework: The Indiamart decision is less about the parties involved and more about the limits of legacy legal categories. It signals that applying a quarter-century-old internet law to AI systems may not be sustainable.
The larger question is not where LLMs fit within the existing framework, but whether that framework needs to evolve. Today’s real regulatory risk may lie in trying to fit AI tools into a legal-liability model that was never designed for them.
The authors are, respectively, partner and head of digital disruption, and associate, Cyril Amarchand Mangaldas.