The AI Hallucination Sanctions Tracker: England & Wales, Singapore and India
Every verified case of unverified AI-generated citations reaching courts in England & Wales, Singapore and India, with the citation, what happened, and the outcome. Updated as new rulings land.
Unverified AI-generated citations have now reached the Supreme Court of India, the High Court of England and Wales, and the General Division of the Singapore High Court. This tracker lists every verified case in the three jurisdictions we cover, with the citation, what happened, and the outcome. We update it as new rulings land rather than writing a fresh article for each one, so bookmark it.
England and Wales
Anthony Malcolm Cork & Anor v Mark Smith [2026] EWHC 1199 (Ch) (High Court, Chancery Division, reported 24 May 2026). In what should have been a routine, uncontentious block transfer application, the applicants’ solicitors told the court that “Rule 12.37(5)” of the Insolvency Rules 2016 granted a power to release a liquidator, quoting wording for it. Judge Mullen checked legislation.gov.uk and a practitioner text directly and found the provision said nothing of the kind. The judge suspected an unchecked AI hallucination and was not persuaded by the solicitors’ subsequent explanation, remarking that legal research and reasoning cannot be outsourced to AI.
R (Ayinde) v London Borough of Haringey, heard with Hamad Al-Haroun v Qatar National Bank QPSC & Anor, [2025] EWHC 1383 (Admin) (Divisional Court, 11 June 2025). Two cases heard together by the President of the Divisional Court. In Ayinde, a judicial review claim included five fake cases in the statement of facts and grounds; the deputy judge had already ordered counsel and the law centre to each pay £2,000 in wasted costs. In Al-Haroun, a claimant seeking £89.4 million in damages cited 45 authorities, of which 18 did not exist, with one fake authority even attributed to the judge hearing the case. The Divisional Court declined to open contempt proceedings in either case but issued a general warning that non-compliance with verification obligations risks “severe sanction,” calling on chambers, firms, and regulators to ensure lawyers understand their duties when using AI.
A registered foreign lawyer, Solicitors Disciplinary Tribunal ruling, 3 September 2026. The SDT banned the lawyer for repeatedly generating inaccurate case citations using AI. This sits alongside the SRA’s own disclosure that it received 42 reports of AI-related misconduct between July 2025 and July 2026, covering inaccurate citations, poor supervision, and confidentiality breaches.
Singapore
Tajudin bin Gulam Rasul and another v Suriaya bte Haja Mohideen [2025] SGHCR 33 (General Division of the High Court, 29 September 2025). Counsel for the claimants cited a case to support an argument under the Moneylenders Act 2008 that did not exist. When challenged, counsel initially described it as a wrong citation and only admitted under questioning that the case did not exist, eventually attributing it to a junior lawyer’s use of an AI app that had produced a “placeholder” citation. The Assistant Registrar found this explanation lacked candour and ordered counsel to personally pay S$800 in costs within 14 days.
Tan Hai Peng Micheal and another v Tan Cheong Joo and another [2025] SGHC 217 (liability finding, 3 November 2025) and the subsequent sanctions ruling of 6 March 2026. In a loan recovery dispute, closing submissions cited two entirely fictitious cases. Justice S Mohan found reasonable grounds to suspect AI generation. In the follow-up ruling, both Goh Peck San and Amarjit Singh Sidhu were ordered to personally pay S$5,000 each (roughly US$3,900), the court rejecting a lower figure the lawyers had proposed and describing the conduct as a “sheer lack of diligence.”
India
Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. & Anr. (Supreme Court of India, reported 2 July 2026). This is the most alarming entry on this list because the AI-hallucinated citations were not introduced by a lawyer. The National Company Law Tribunal, Mumbai, had admitted insolvency proceedings against Essel Infraprojects Ltd on 28 August 2024, and the National Company Law Appellate Tribunal upheld that decision on 11 September 2025. The Supreme Court found that the NCLT itself had relied on fabricated judgments and genuine citations carrying invented passages, introduced through the tribunal’s own research rather than counsel’s submissions. A bench of Justices Pamidighantam Sri Narasimha and Alok Aradhe set aside both orders as “tainted” and restored the original proceedings for expedited rehearing. The Court called for a “zero-tolerance mode” toward AI-generated precedents and directed the Bar Council of India to form a committee on disciplinary guidelines.
Vijay Ghanshyam Gadiya v. Union of India & Anr., 2026 INSC 947 (Supreme Court of India, 2 September 2026). A customs authority in Surat had imposed a ₹425.27 crore penalty for alleged mis-declaration of diamonds, upheld by the Gujarat High Court on 20 January 2026. Before the Supreme Court, Justices Dipankar Datta and Sheel Nagu found that the customs order itself relied on citations that did not exist and genuine judgments misquoted with fabricated passages, an apparent AI hallucination. The Court set aside both the customs order and the Gujarat High Court’s ruling, remanding the matter to a different officer, and described AI as “training wheels,” not the “pilot’s seat.” It expressly relied on Pooja Ramesh Singh as precedent for zero tolerance.
The “Mercy vs Mankind” filing (Supreme Court of India, 17 February 2026). A bench led by CJI Surya Kant, with Justice BV Nagarathna, confronted a petition that cited a case called “Mercy vs Mankind” as binding authority. It does not exist. The bench also flagged genuine Supreme Court judgments cited with quotations that do not appear in them. No individual sanction has been reported from this specific hearing, but it is widely credited with accelerating the Supreme Court’s draft Regulations for Use of Artificial Intelligence in Courts, 2026, published for consultation on 3 June 2026.
What the Pattern Actually Shows
Three jurisdictions, three different court systems, and the same two failure points every time: nobody opened the actual case before filing it, and nobody expected the court to check. Only one case on this list, Pooja Ramesh Singh, involves a tribunal’s own unchecked research rather than a lawyer’s submission, which is precisely why it drew the sharpest judicial language of any case here.
Several Singapore firms, including TSMP Law, Withers KhattarWong, Drew & Napier, and Allen & Gledhill, have since treated AI policy breaches as disciplinary matters with consequences ranging from retraining to dismissal. That is the direction every jurisdiction on this list is heading, whether or not formal rules exist yet.
This is exactly the discipline we teach in our Legal Research module: never let an AI-generated citation into a filing until a human has opened the source and confirmed it says what you think it says. Every case above started with someone skipping that one step.