Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. & Anr., 2026 INSC 668

Pan India Utilities Distribution Company Ltd. (PIUDCL) borrowed from Jammu and Kashmir Bank, with a corporate guarantee from Essel Infraprojects Ltd. (EIL). When PIUDCL defaulted and its accounts became non-performing assets, the Bank applied under Section 7 of the Insolvency and Bankruptcy Code, 2016 to start the corporate insolvency resolution process against EIL as guarantor. Bar & Bench reported the claimed debt as Rs 87.43 crore arising from a Rs 200 crore facility secured by the guarantee and a mortgage over land at Gorai, Borivali.¹³

On 28 August 2024 the NCLT, Mumbai admitted the application, appointed an interim resolution professional and declared a moratorium. EIL, through its suspended director Pooja Ramesh Singh, argued before the NCLAT that its liabilities had moved to other companies through a demerger and later amalgamation, and that a renewed sanction letter of 18 November 2017 omitted the guarantee. The NCLAT dismissed the appeal on 11 September 2025, holding that internal restructuring could not affect the guarantor's liability, relying on clause 8 of the guarantee deed.¹

In the Supreme Court, Senior Advocate Madhavi Divan pointed out at the outset that the citations relied on by the NCLT to reach its findings, as reproduced in paragraph 12 of the NCLAT's opinion, were "fake and non-existent, probably AI-generated", and that even where case citations were accurate the excerpted paragraphs could not be traced in the law reports.¹

The Court's own check, set out in paragraph 15, was as follows.

The Bank filed an affidavit stating that its counsel had not cited these judgments and that the precedents relied on by the NCLT were "obtained through its own research". The Court also noted that the fake judgments "escaped scrutiny by the first statutory appellate tribunal", and observed that courts and tribunals "implicitly trust lawyers when referring to precedents" and that it would be impractical for a court to verify every judgment an advocate cites.¹

Justice Narasimha opened by calling this "yet again a case where the Tribunal relied on non-existent, fake and hallucinated material, generated through Artificial Intelligence". The judgment places the case inside a larger argument. AI "has acquired the capability to better, if not fully substitute, human effort, both routine and intellectual", and professionals are being compelled to adopt it, but "if left unregulated" it may "infiltrate our intellectual work ethic and, before long, render us dependent on its vast capabilities". Courts have absorbed earlier technologies as aids, whereas AI "is not just an aid to assist us in our work, but is an alternative to our own thinking, reasoning and even decision making".¹

The Court then separated two questions. Regulation of AI use is a matter for public policy and "enforceable Rules and Regulations", which it noted were being deliberated, and not for judicial declaration. But the legal position on hallucinations could be stated. The Court was "neither concerned with the cause nor with the process of resolving such hallucinations", which it left to engineers and scientists, and focused on fake material's use "as precedents in law". Paragraph 7 sets the rule: zero tolerance for the Bar and the Bench, misconduct for an advocate who cites unverified AI precedents, a serious lapse for a judge who relies on them, and nullity, "no decision in the eyes of the law", for any decision tainted by them, "irrespective of whether such material had a direct or indirect bearing". The Court added that its ruling "shall have no bearing on the rightful use of AI".¹

Paragraph 8 insists that "mere declaration of prohibitory action is not sufficient; there must be a consequential action following accountability". Paragraph 17 adds that a decision based on fake or hallucinated material "amounts to subversion of the rule of law". In a footnote the Court cited an English case in which a law firm's AI programme produced a fabricated statutory quote, the judge declined contempt proceedings but publicly admonished the firm, and the firm referred itself to its regulator.¹

The Court set aside the NCLT order of 28 August 2024 and the NCLAT judgment of 11 September 2025, restored the Section 7 application to its original number (RCP (IB) 6/MB/2023), and directed the NCLT to decide it on its own merits, preferably within two weeks, with status quo maintained in the meantime. It expressed no opinion on the merits.¹

The separate direction to the Bar Council of India is the main regulatory consequence. Describing the Council as "the apex statutory body", the Court told it to constitute a committee, deliberate "with utmost seriousness", "prescribe a guiding principle to prevent such occurrences, along with the disciplinary action that will follow a violation of the norms".¹² This followed a related matter, Gummadi Usha Rani v. Sure Mallikarjuna Rao, in which the same Bench took cognisance on 27 February 2026 of a trial court's reliance on AI-generated judgments,⁸ and on 5 May 2026 orally asked the Council to set up an expert committee, observing that the AI Committee of the Supreme Court was "almost at the verge of finalising the draft regulations".⁶

The judgment is the Supreme Court's most developed ruling so far on AI hallucination in adjudication, and its language has already been taken up. On 2 September 2026 a Bench of Justices Dipankar Datta and Sheel Nagu relied on it to set aside a Rs 425.28 crore customs penalty order after finding that the adjudicating officer had relied on case law that was non-existent or carried fake citations.⁴ In July 2026 the Supreme Court Advocates-on-Record Association invoked the case in its comments on the draft Regulations for Use of Artificial Intelligence in Courts, 2026 to argue that AI output must remain subject to meaningful human oversight, while seeking removal of a mandatory AI-disclosure requirement for lawyers and proposing instead a certificate of personal verification of citations.⁵

Several questions remain open, and the judgment does not address them. The rule that "an iota" of fake material vitiates a decision is stated without a materiality threshold, so how it applies to a trivial mis-citation that did not influence the result is unclear. The Court acknowledged that judges cannot verify every cited authority, but did not say what verification duty tribunals themselves bear. The direction to the Bar Council leaves the content of the guiding principle and sanctions to that body, and the sources reviewed did not show that it has yet published norms. The Court also declined to define hallucination technically, leaving that to engineers. The draft national regulations would add a disclosure duty for AI-assisted filings (Regulation 43(3)) and a rule that a person whose filing is fabricated or misleading by reason of its AI-generated character cannot rely on that as a defence (Regulation 43(6)).⁷

As of 15 October 2026 the appeal is disposed of and the Section 7 application stands restored before the NCLT, Mumbai. The sources reviewed did not show the outcome of the fresh NCLT hearing. The Bar Council of India's committee had not, in those sources, published its guiding principle or disciplinary norms, and the Supreme Court's draft AI regulations remained a draft.