Abstract
Generative artificial intelligence has introduced a different evidentiary issue into legal practice i.e. fluent and entirely fictitious case law. Unlike manipulated documents or deepfakes which can be easily exposed through forensic or technical tools, a hallucinated citations carriers no visible defect. It looks exactly like authentic precedent until changed against a verified source. This article will examine how Indian courts have responded to the fabricated precedent in pleadings and even judgement themselves. The article argues that the difficulty here is not technological detection but professional discipline under pressure and India’s existing legal framework under the Advocates Act, 1961 doesn’t yet introduce the supreme court’s zero tolerance pronouncement and day to day verification standard for the profession.
Keywords
Generative artificial intelligence, hallucinated precedent, legal ethics, professional misconduct, judicial Verification, Bar Council Of India, Advocates Act,1961
Introduction
The most consequential legal risk posed by AI in day to day practice in not deepfake or manipulated documents, it is something far more dangerous i.e. a case citation that exactly looks correct, reads with the rhythm of a real judgement but doesn’t exist. A lawyer confronting a forged or morphed photograph or an altered recording can be checked through forensic tools but a lawyer confronting a fabricated citation embedded in a brief or a judge relying on one embedded in a own research, has only one reliable defence i.e. opening verified database and checking every word.
The Anatomy of the problem:
Artificial intelligence models are built to produce statistically credible text not to retrieve verified records. When prompted for a case law , a model may generate a citation with a realistic party name , a credible court, a coherent year and an internally consistent quotation – All these doesn’t bear the authenticity that alert a reader to forgery. AI manipulated evidence can be tested against forensic markers id manipulation however a hallucinated citations has no such market. It is not altered and never existed. The only test available is cross referencing against an authoritative database and the test entirely depends on human diligence. This is why the problem persists despite two years of judicial warnings. It is not a detection problem awaiting for a better tool , but a verification discipline problem occurring due to time pressure, docket backlog and overconfidence in AI output combine.
The Doctrinal Response: Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd.
On 2nd July 2026 , the Supreme Court in an appeal arising from a insolvency proceedings under Section 7 of the Insolvency and Bankruptcy Code, 2016 , discovered that the National Company Law Tribunal and on appeal, the National Company Law appellate Tribunal had relied on precedents that were wholly non existent or genuine citations padded with fabricated paragraphs which were never part of the original judgement. Notably, the fabricated material had not been introduced by counsel but originated from the Tribunal’s own research. The court set aside both orders and stated “ a decision of a court or an adjudicating authority based on material which is fake and hallucinated is no decision at all, and it amounts to subversion of the rule of law. The Bench declared that citing such material without verification is misconduct in the part of an advocate, that an equally serious lapse arise where a judge relies on such material and that a decision is to be set aside even if an iota of fake or hallucinated material enters the decision making process.” The court directed the Bar Council of India ton constitute a committee to frame guiding principles and disciplinary consequences for advocates who place such material before courts.
Institutional Failures Across Tiers
The case of Pooja Ramesh Singh didn’t arise in isolation. In December 2024, the Bengaluru bench of the Income Tax Tribunal recalled a ruling in a matter involving a trust with an approximately 669 crore tax dispute after discovering that it’s own order rested on four fabricated citations. In October 2025, Bombay High Court quashed a roughly 28 crore faceless tax assessment after finding that the assessing authority had relied on three non existent judicial decisions observing that reliance on AI generated results can’t be blind particularly where quasi judicial functions are being exercised. Weeks before the Supreme Court’s July 2026 ruling, it had already taken suo muto cognisance of an Andhra Pradesh trial court’s reliance on four fabricated Supreme Court judgements, declaring such reliance misconduct warranting legal consequences. The pattern is intrusive precisely because it implicates every level of system i.e. trial courts, Tribunals and appellate benches and because in several instances the fabricated material originated from judicial or Tribunal research rather than from party’s submissions it shows that the risk is not confined to the private bar .
Comparative Perspective
The United States confronted this problem earlier and has generated a corresponding more developed sanctions jurisprudence. In Mata v. Avianca, Inc. , the southern District of New York sanctioned two attorneys and their firm under Federal rule of Civil Procedure after they submitted a brief citing six entirely fabricated cases generated by ChatGPT . The Judge found ‘subjective bad faith ’ sufficient for sanctions. In Johnson v. Dunn, the court concluded that monetary sanctions alone had proved insufficient to deter hallucinated citations and instead disqualified the responsible attorneys from continued representation in the matter, directing that the opinion be published and reported to state bar regulators. The trajectory across jurisprudence is consistent, in initial phase a modest monetary sanction of recurrence continued it goes to disqualification and referral to disciplinary bodies.
Critical Analysis
The Advocates Act, 1961 and the Bar Council of India rules were framed decades before generative AI existed and none of the instrument presently articulates a specific verification standard for AI assisted legal research. The case of Pooja Ramesh Singh announces a principle but leaves it’s operational content to a committee yet to be constituted. Secondly, the risk is not confined to counsel, the Supreme Court’s own findings show tribunals conducting AI assisted research without disclosure and without a verification protocol. Furthermore, the trial courts and tribunals facing extraordinary docket pressure, with more than 5.29 crore cases pending across Indian Judiciary which are more likely to learn on AI assisted drafting and least likely to have the time to verify each citation against a primary source. The Court’s own white paper on Artificial intelligence and the judiciary, released in November 2025, had already warned this exact risk without prescribing a binding accountability mechanism which persisted until Pooja Ramesh Singh forced the issue five months later.
Recommendations
First the Bar Council of India should adopt a specific binding rule under the Bar Council of India Rules requiring advocates to certify at the foot of any pleading citing case law that each citation has been independently verified against an authorised law report or database. Secondly and equivalent verification protocol should be extended administratively to judicial and tribunal research staff to check the Court’s own findings that fabricated material had entered through institutional rather than party research. Thirdly, the judiciary should mandate disclosure requirements comparable to standing orders adopted by several United States District Courts. Fourthly, the courts should adopt a graduated sanctions framework, moving from cost orders on a first instance to referral for disciplinary action. But the Indian experience over the past two years shows that warnings and models costs alone have not stopped the pattern .
Conclusion
The Supreme Court’s ruling in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. Marks a genuine turning point that hallucinated precedent is misconduct rather than an error and decision by it is jo decision in the eyes of law. Yet the harder problem the case exposes is not Doctrinal but behavioural. A manipulated recording, a fabricated citations offers no forensic detection, it can only be detected by checking every citation against a verified source like SCC Online, Manupatra or Indian Kanoon etc. Until India’s professional conduct framework implements the Supreme Court’s zero tolerance policy into a specific, binding and institution wide verification standard, the pattern is likely to recur.
THIS ARTICLE IS WRITTEN BY Y JOHN SAMUEL FROM SAMBALPUR UNIVERSITY
REFERENCES :
iPleaders, AI-Hallucinated Case Law: How Fake Citations Are Getting Lawyers Sanctioned in India (and How to Avoid It) (2026), https://blog.ipleaders.in/ai-hallucinated-case-law-fake-citations-india/.
LiveLaw, Phantom Precedents: The Rise of AI-Generated Case Law in Indian Courts (2026), https://www.livelaw.in/articles/phantom-precedents-ai-generated-case-law-indian-courts-526665.
Legal Service India, Actual Intelligence over Artificial Intelligence: Impact of Hallucinated AI-Generated Citations in Written Submissions and Judgments (2026), https://www.legalserviceindia.com/Legal-Articles/indiscriminate-use-of-ai-in-legal-submissions-hallucinated-case-laws/.
Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 448 (S.D.N.Y. 2023).
MediaNama, SC Declares AI-Generated Fake Citations “Misconduct”: What Does This Mean for India’s Judiciary? (Mar. 3, 2026), https://www.medianama.com/2026/03/223-supreme-court-ai-fake-case-laws-misconduct/.
Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., 2026 INSC 668, ¶ 1 (India).
The Indian Lawyer, When Technology Crosses the Line: Safeguarding the Sanctity of Judicial Decision-Making (2026), https://theindianlawyer.in/when-technology-crosses-the-line-safeguarding-the-sanctity-of-judicial-decision-making/.
Bombay High Court order (Oct. 2025), quoted in Bar and Bench, Bombay High Court Pulls Up Tax Officer for Citing Fake AI-Generated Case Laws, https://www.barandbench.com/news/litigation/bombay-high-court-pulls-up-tax-officer-for-citing-fake-ai-generated-case-laws.
Verdictum, Zero-Tolerance for Using AI-Generated Precedents Without Verification: Supreme Court Sets Aside NCLT, NCLAT Orders Citing Fake Judgments (2026), https://www.verdictum.in/supreme-court/pooja-ramesh-singh-v-jammu-and-kashmir-bank-ltd-2026-insc-668-ai-generated-precedents-1616998.
Johnson v. Dunn, No. 2:21-cv-1701 (N.D. Ala. July 23, 2025), discussed in Esquire Deposition Solutions, Federal Court Turns Up the Heat on Attorneys Using ChatGPT for Research (2025), https://www.esquiresolutions.com/federal-court-turns-up-the-heat-on-attorneys-using-chatgpt-for-research/.
Anthony Malcolm Cork v. Mark Smith, [2026] EWHC 1199 (Ch), discussed in Law Web, Supreme Court: Decision Based on Fake Precedents Are Void (2026), https://www.lawweb.in/2026/07/supreme-court-decision-based-on-fake.html.
Advocates Act, 1961, §§ 35, 49 (India);
Supreme Court of India, Centre for Research and Planning, White Paper on Artificial Intelligence and the Judiciary (Nov. 2025), discussed in MediaNama, supra note 7.
Barristery, Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. 2026 (2026), https://www.barristery.in/2026/07/pooja-ramesh-singh-vs-jammu-and-kashmir-bank-2026.html.
Johnson v. Dunn
