The Supreme Court's AI-Hallucination Warning: What Indian Advocates Need to Change in Their Research Workflow

Pooja Ramesh Singh, a suspended director of Essel Infraprojects Ltd. (EIL), challenged proceedings arising from EIL's role as corporate guarantor for credit facilities that Pan India Utilities Distribution Company Ltd. (PIUDCL) had availed from Jammu and Kashmir Bank Ltd. The matter reached the Supreme Court after passing through the National Company Law Tribunal (NCLT) and the National Company Law Appellate Tribunal (NCLAT).

The issue that made this case significant had nothing to do with the underlying guarantee dispute: the NCLT's order relied on material - presented as legal precedent - that had been generated by AI and did not correspond to any real, existing judgment.

Court: Supreme Court of India. Bench: Justices P.S. Narasimha and Alok Aradhe. Case: Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. & Anr., Civil Appeal No. 11950 of 2025. Neutral citation: 2026 INSC 668. Decided: 2 July 2026.

The Court set aside both the NCLT's order and the NCLAT's appellate order, holding that an adjudicatory order cannot stand if it relied on fabricated, AI-hallucinated material presented as precedent - the integrity of the adjudicatory process itself was compromised, independent of how the underlying guarantee dispute might otherwise have been decided.

The Court went further than resolving this one case: it directed the Bar Council of India to constitute a committee to examine advocates submitting AI-generated fake or hallucinated material as if it were genuine legal precedent, and to prescribe guiding principles - including a disciplinary framework - to address the practice going forward.

This is not a ban on using AI tools in legal research or drafting. The judgment addresses fabricated material being relied upon as if genuine - it does not hold that AI-assisted research is impermissible when the output is actually verified before use.

It is also not, by itself, a finished disciplinary framework - the Court directed the Bar Council of India to build one; that committee process is prospective, not something this judgment completes on its own. Advocates should watch for what the BCI committee actually produces rather than assuming a specific rule already exists beyond what this judgment states.

This case gives the pattern documented in our companion piece on AI-hallucinated citations in Indian courts its most consequential outcome yet: it is no longer just trial-court orders or tribunal findings being corrected - an NCLT order was set aside entirely by the Supreme Court over it, with a direct institutional response aimed at the profession.

For any firm or advocate using AI tools in research or drafting, the practical takeaway is the same one this case makes unavoidable: every citation used before any tribunal or court needs to be independently confirmed against a primary source before it's relied upon - not because a specific rule requires it yet, but because an adjudicatory order built on an unverified citation is now demonstrably at risk of being set aside entirely, not just corrected.

At minimum: confirm the case actually exists on a primary source (the court's own website, NJDG, or a reliable database); confirm the court, date, and citation identifier match; and confirm the cited paragraph actually supports the proposition it's being used for. See our companion article for a fuller ten-point verification framework built around exactly this kind of failure.

This judgment is a concrete, high-stakes illustration of the exact problem source-grounded AI is meant to address: an AI system whose output is tied to an identifiable, checkable source lets an advocate confirm a citation before it reaches a filing, rather than discovering the problem after an order has already been set aside.