Calcutta HC Rules ChatGPT an Originator, Raises Liability Questions

The Calcutta High Court considered Indiamart Inter Mesh Ltd v OpenAI Inc and concluded, on a prima facie basis, that ChatGPT's generative functions suggest it may be an originator rather than an intermediary, a question the court described as "complicated and vexed" (Calcutta High Court judgment dated 20 May 2026; reported by SCC Online and indiankanoon). The court dismissed IndiaMart's application for interim relief that alleged selective discrimination and diversion of business by ChatGPT, as recounted in reporting by LiveMint, CyberPeace, and MediaNama. The May 20 order reverses the court's December 2025 preliminary finding, in which the same judge had found a "strong prima facie case" of selective discrimination and said IndiaMart was excluded "without any logic"; after hearing OpenAI's full defence, that position did not hold (MediaNama). India has approximately 100 million weekly ChatGPT users, the second largest user base globally, giving this ruling substantial weight for AI platform governance in emerging markets.

Industry reporting and the judgment summary record that ChatGPT and similar large language models generate synthesized answers rather than returning a ranked list of third-party links, which is the functional distinction at the centre of the dispute (LiveMint; CyberPeace). The petitioner asserted that ChatGPT displayed links to competing services while omitting IndiaMart listings; IndiaMart also alleged that rivals named on the USTR Notorious Markets List -- DHGate, Pinduoduo, Shopee, Taobao -- continued appearing on ChatGPT, making the exclusion selectively discriminatory. The court noted IndiaMart had previously blocked ChatGPT from crawling its website and used that against the company's harm argument.

Editorial analysis - technical context: Generative models produce outputs by probabilistic synthesis from trained weights and training data rather than by serving stored third-party content. Observed patterns in similar litigation internationally show courts confronting whether an AI output is attributable to the platform as original speech or to underlying data providers. For practitioners, this distinction affects how takedown obligations, notice-and-takedown mechanics, and content-moderation workflows are framed when AI systems create rather than merely link to content.

India's intermediary safe-harbour under Section 79 of the Information Technology Act, 2000 was drafted for a web where platforms primarily hosted or transmitted third-party material; multiple commentaries argue the provision rests on a conduit model that generative AI challenges (LiveMint; SCC Online). The IndiaMart judgment is being read across legal and policy outlets as an early test case for whether existing liability shields can be applied to systems that actively generate answers, not simply relay links (CyberPeace; Ikigai Law; MediaNama).

Observers will track whether the main suit proceeds and how the court addresses evidentiary questions about model architecture, training data, and output provenance at a full hearing. Industry stakeholders and regulators drafting amendments to the IT Rules and AI-specific governance frameworks will likely cite this ruling when debating whether intermediaries that produce synthesized responses require new disclosure, auditability, or attribution duties. For downstream teams, technical signals to monitor include documentation of source attribution layers, logging of prompt-output provenance, and controls for surfaced links in AI-augmented search features.