On 2 July 2026, a Division Bench of the Supreme Court of India, comprising Justices Pamidighantam Sri Narasimha and Alok Aradhe, set aside the orders of the National Company Law Tribunal, Mumbai (“NCLT”) and the National Company Law Appellate Tribunal (“NCLAT”) in Pooja Ramesh Singh v Jammu and Kashmir Bank Ltd, after finding that the adjudicating authority had relied upon fabricated, non-existent and artificial-intelligence-hallucinated case law while admitting a company into insolvency.
The ruling, reported as 2026 INSC 668, is the first occasion on which the Supreme Court has laid down a binding, “zero-tolerance” standard for the citation and judicial use of artificial intelligence-generated material. Its reasoning has immediate consequences not merely for insolvency practice under the Insolvency and Bankruptcy Code, 2016 (“IBC”), but for the integrity of adjudicatory processes across company law, arbitration and constitutional litigation more generally.
I. Factual Matrix
The appeal arose out of insolvency proceedings initiated by Jammu and Kashmir Bank Ltd under Section 7 of the IBC against Essel Infraprojects Ltd (“EIL”), which had furnished a corporate guarantee for credit facilities extended to Pan India Utilities Distribution Company Ltd. Following default by the principal borrower, the bank invoked the guarantee and moved the NCLT for admission of EIL into the corporate insolvency resolution process. The appellant, a suspended director of EIL, challenged the NCLT’s admission order before the NCLAT and, upon its affirmance, before the Supreme Court under Article 136 of the Constitution.
During the hearing, senior counsel for the appellant demonstrated that several of the “precedents” relied upon by the NCLT while admitting the Section 7 application did not correspond to any authentic judicial decision. Two citations in particular, purportedly ICICI Bank Ltd v Urban Infrastructure Real Estate Ltd (2019) 16 SCC 528 and Sarbjit Singh v Union Bank of India (2022) 7 SCC 464, were found on verification to be entirely non-existent, while certain other extracts contained paragraphs falsely attributed to genuine reported decisions. Neither the NCLT nor the NCLAT had detected the fabrication before the matter reached the Supreme Court.
II. The Court’s Reasoning
The Bench held that reliance on unverified, AI-generated legal authorities strikes at the root of adjudicatory legitimacy, irrespective of whether the fabricated material was outcome-determinative. The Court observed that a decision founded even partly on hallucinated precedent “is no decision in the eyes of the law” and must be set aside “even if an iota of fake or hallucinated material enters the decision-making process,” since its presence “would violate the sanctity of adjudication”.
This is a materially more exacting standard than the ordinary appellate inquiry into whether an error affected the outcome; the Court treated the mere entry of fabricated material into the reasoning process as fatal, regardless of causal weight. The Bench was careful, however, not to condemn the use of artificial intelligence as a research aid outright. The Court simultaneously endorsed the use of AI tools to assist adjudication, while insisting that judicial decision-making must remain under “total and absolute” human control. The distinction drawn is between AI as an unverified source of authority and AI as a tool whose output is independently checked against primary sources before being relied upon – a distinction with clear resonance for legal publishing and research practice.
III. Attribution of Responsibility and the Role of the Bar Council
Significantly, the Court did not confine its censure to the tribunal. It held that citing unverified AI-generated judgments is professional misconduct on the part of an advocate, and an equally serious lapse where a judge relies on such material in support of a determination. The Court directed the Bar Council of India to constitute a committee to frame guiding principles on the use of artificial intelligence in litigation and to prescribe disciplinary consequences for advocates who place fabricated authorities before courts. This is a notable extension of judicial supervisory power under Article 142 of the Constitution into the domain of professional regulation, and it places Indian courts among the first in the world to translate concerns about generative AI hallucinations into an enforceable, institution-wide compliance obligation on the profession rather than a caution addressed to individual counsel. Whether this sits comfortably with the Court’s own description of Article 142 as a “supplementary” power that cannot be used “to build a new edifice where none existed” is a question the profession is likely to test as the Bar Council’s committee begins its work.
IV. Implications for Company and Insolvency Law Practice
The immediate practical consequence of the ruling is procedural: NCLT and NCLAT orders founded on unverifiable citations are now vulnerable to being reopened on this ground alone, without the appellant having to independently establish that the outcome would have differed absent the fabricated material. Given the volume of insolvency litigation currently before the tribunals, and the increasing use of AI-assisted drafting by counsel and, on the tribunals’ own admission, by benches under time pressure, the judgment effectively mandates a verification protocol before final orders are signed. The Court itself acknowledged that this cannot be achieved through judicial orders alone, noting that the necessary safeguards would follow through public policy and enforceable Rules and Regulations that are already under deliberation.
For company law practitioners more broadly, the decision sits alongside a wider 2026 trend of the Supreme Court tightening procedural discipline in tribunal adjudication – including its clarification, in Dr Bais Surgical and Medical Institute Pvt Ltd v Dhananjay Pande, that the register of members is not a precondition for maintaining oppression and mismanagement proceedings under Sections 397–398 of the Companies Act, 1956 (Now, Sections 241-242 of the Companies Act, 2013). Read together, these rulings reflect a Court increasingly attentive to the procedural integrity of company-law adjudication at the tribunal level, whether the defect concerns the substantive threshold for relief or, as here, the authenticity of the legal material underlying the decision itself.
V. A Note of Caution for Legal Research and Publishing
The judgment has an evident bearing on legal writing and publishing practice, an audience of direct relevance to this platform. Researchers relying on generative tools for case discovery should never cite an AI-generated case reference directly, and must cross-verify the case name, citation and operative paragraphs against an authenticated database – the official website of the Supreme Court or the relevant High Court, or established repositories such as SCC Online.
The Court’s own summary of the ruling is available on the Supreme Court’s official landmark judgment summaries page, which itself carries the caveat that such summaries are prepared for general understanding and do not substitute the original record – a fitting reminder, given the subject matter of the case, that even authorised secondary material should not be treated as a substitute for verification against the primary judgment.
The comparison with other jurisdictions is instructive. United States courts confronting the same problem, most notably in Mata v Avianca, Inc, where counsel who submitted ChatGPT-fabricated citations were sanctioned, have proceeded case by case, treating each instance as a matter of individual lawyer discipline. The Indian response, by contrast, combines judicial invalidation of the tainted decision itself with a direction to the Bar Council of India to frame prospective, institution-wide norms for the profession as a whole, a structurally broader intervention than the sanctions-based approach taken abroad.
VI. Conclusion
Pooja Ramesh Singh does more than resolve a dispute over a corporate guarantee and a Section 7 admission order. It articulates, for the first time with binding force, the proposition that the presence of fabricated authority in the chain of judicial reasoning is an independent and sufficient ground of invalidity, and it imposes on both the Bench and the Bar a continuing duty of verification.
As artificial intelligence tools become embedded in litigation practice and in the tribunals themselves, the case is likely to be cited with increasing frequency – not only in insolvency matters, but wherever a party can show that unverified AI output found its way into the record of adjudication. For a profession that increasingly turns to generative tools as a first port of call in research, the judgment’s real instruction is a return to first principles: authority in law is only as good as its source, and no technological convenience relieves counsel or the court of the obligation to check it.