When AI Enters the Courtroom, Verification Becomes Justice

Every courtroom runs on an assumption that is rarely stated aloud but always relied upon: when a judgment is cited, it exists. Not as description or approximation, but as a real, traceable legal record. That quiet assumption is what gives legal reasoning its authority. When it breaks, even in a single instance, the damage goes beyond one case. It strikes at the credibility of the process itself.

That fragility was on full display on July 2, 2026, when the Supreme Court of India delivered its judgment in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. & Anr. While examining an insolvency dispute, a bench of Justices P.S. Narasimha and Alok Aradhe found that the National Company Law Tribunal had relied on judicial precedents that did not exist in any recognised legal record. These were not misread cases or interpretative errors. They were AI fabricated citations, treated as if they were valid law.

Also Read:  ONOE: The Reform That Cheapens Your Vote and Cuts Your Mandate

A Verdict That Named the Problem

The case began as a routine corporate insolvency matter. Jammu and Kashmir Bank had moved the NCLT under Section 7 of the Insolvency and Bankruptcy Code against Essel Infraprojects Ltd., seeking recovery of dues owed by the original borrower, Pan India Utilities Distribution Company Ltd. The NCLT admitted the insolvency application, and the NCLAT upheld that order by relying on a string of what it presented as Supreme Court precedents. When the matter reached the Supreme Court, the appellant, a suspended director of Essel Infraprojects, argued that several of those precedents simply did not exist.

On independent verification, the Court agreed. It found citations such as ICICI Bank Ltd. vs Urban Infrastructure Real Estate Ltd. and Sarbjit Singh vs Union Bank of India, both attributed to the Supreme Court Cases reporter, to be entirely fictitious, while other citations carried paragraphs and quotations that never appeared in the genuine judgments they claimed to come from.

Jammu and Kashmir Bank filed an affidavit clarifying that its own lawyers had never placed these cases on record, and that the tribunal appeared to have generated them through its own research. The Supreme Court held that the source of the error did not lessen the damage done to the rule of law, and it set aside the NCLT and NCLAT orders, sending the insolvency application back for fresh adjudication.

The bench did not mince words about what such material does to judicial reasoning once it enters the system unchecked, comparing its spread to a toxic industrial leak: invisible at first, and catastrophic by the time anyone notices. The Court declared that courts must adopt a zero-tolerance approach to producing, citing, or relying on AI generated precedents without verification, and it went further, holding that a lawyer who cites unverified machine-generated judgments commits professional misconduct, while a judge who relies on such material commits an equally serious lapse. The Court also directed the Bar Council of India to constitute a committee to frame guidelines and consider disciplinary measures for the use of unverified, unchecked machine-generated material in legal work.

Also Read:  The CPA Scandal That Exposed Delhi's Broken Procurement System

Not India's First Encounter, and Not the World's

India's judiciary has already brushed up against this problem more than once this year. Earlier in 2026, the same bench had dealt with a separate matter in which a trial court had relied on non-existent, AI generated judgments, a recurrence that the Court itself flagged as reason enough to move toward firmer standards.

In April 2026, the Punjab and Haryana High Court went further still, issuing a circular related to AI that bars judicial officers from using generative tools such as ChatGPT, Gemini, and Copilot for legal research or the writing of judgments, citing data-privacy concerns under India's yet-to-be-enforced Digital Personal Data Protection Act, 2023.

The problem is neither new nor confined to India. The first widely reported instance came from the United States, in Mata v. Avianca, Inc., decided in the Southern District of New York in 2023. Two lawyers representing a passenger who claimed he was injured by a serving cart on an Avianca flight used ChatGPT to research their opposition to a motion to dismiss. The tool invented six complete judicial opinions, complete with fabricated judge names, docket numbers, and quotations.

When opposing counsel could not locate the cases, one of the lawyers reportedly asked the chatbot to confirm the cases were real, and it did, itself hallucinating a second time. Judge P. Kevin Castel fined the lawyers and their firm 5,000 dollars, not for using AI as such, but for standing behind the fabricated cases even after being warned they might not exist.

The United Kingdom has logged dozens of comparable episodes. In the combined 2025 High Court matters of R (Ayinde) v. Haringey and Al-Haroun v. Qatar National Bank, judicial assistants checking one set of submissions found that eighteen of forty-five cited cases did not exist at all, while several genuine cases were quoted for propositions they never contained.

The presiding judges referred the lawyers involved to their professional regulators and warned that fabricated citations amount to an abuse of the court's process, regardless of whether the person responsible is a qualified lawyer or an unrepresented litigant. A related case, Bandla v. Solicitors Regulation Authority, saw an appeal struck out entirely after a former solicitor kept relying on fake authorities even after being told they were fake.

Also Read:  Patent Filings in India Surge—Are Universities Faking Innovation?

The Scale of the Problem, in Numbers

What began as a scattering of isolated incidents has become a measurable pattern. Damien Charlotin, a Paris-based legal researcher affiliated with HEC Paris's Smart Law Hub, maintains the most widely cited public database of court and tribunal decisions in which generative AI produced fabricated legal material. By mid-2026, that database had logged more than 1,400 cases worldwide, more than 1,000 of them in the United States alone, with new entries arriving, in Charlotin's own description, at a pace of roughly ten cases from ten different courts on a single day.

Penalties have escalated alongside the volume: from the original 5,000-dollar fine in the Avianca matter to sanctions exceeding 100,000 dollars in a 2026 case out of Oregon, and to what appears to be the first indefinite licence suspension tied to AI-fabricated citations, handed down by the Nebraska Supreme Court in April 2026.

Most of the cases in Charlotin's tracker involve self-represented litigants rather than practising lawyers, which is its own warning sign: AI tools are now a first stop for people who cannot afford legal counsel, and courts are having to build new safeguards to protect litigants from the very tools meant to help them navigate the system alone. But it is the cases involving licensed advocates that carry the heaviest professional consequences, ranging from wasted-costs orders and referrals to bar regulators, to strike-outs of entire pleadings and, increasingly, suspension from practice.

Courts Are Building Guardrails, Not Banning the Technology

None of this amounts to a rejection of the technology. The Supreme Court's AI Committee released a draft framework, the Regulations for Use of Artificial Intelligence in Courts, 2026, in June, opening it for public comment, with the consultation window extended to July 15, 2026, before the rules are finalised. The draft permits the use of artificial intelligence for legal research, citation verification, drafting assistance, translation, transcription, and court administration, but it draws a hard line: such systems cannot decide cases, assess bail eligibility, evaluate flight risk, predict recidivism, or judge the credibility of a witness or party.

The draft also proposes a mandatory disclosure requirement, meaning any lawyer or litigant who uses such tools to prepare a pleading or submission would have to say so at the time of filing, and it envisages a permanent apex oversight body at the Supreme Court, backed by dedicated technology committees and annual audits of any such system used in Indian courts. Roughly 53.57 crore rupees has separately been earmarked under Phase III of the government's e-Courts Project for future technological advancement, including artificial intelligence and blockchain integration, according to a Lok Sabha reply by the Union Law Minister in December 2025.

This mirrors the direction courts have taken elsewhere. UK judges operate under a formal Guidance for Judicial Office Holders on the use of artificial intelligence, first issued in December 2023, and the Law Society has published its own checklist for practitioners weighing whether and how to use generative tools. 

In the United States, the American Bar Association issued its first formal ethics opinion on lawyers' use of generative tools in July 2024, making clear that the duties of competence, candour to the court, and client confidentiality apply in full regardless of whether a human or a machine drafted the first version of a document.

Also read:  RBI Bans Dark Patterns: A Confession Dressed as a Circular

The Deeper Fix Starts Earlier Than the Courtroom

The challenge, though, is not confined to courts or regulators. It begins earlier, in legal education itself. Law students already use AI tools widely for summaries, notes, and early-stage research, and that is neither unusual nor inherently a problem. The concern arises when these tools replace verification rather than assist it. Legal training has always rested on one core discipline: nothing is accepted as law unless it can be traced back to an authentic, checkable authority. If that habit weakens early, it rarely returns in full strength later in practice.

Practical guidance is converging on similar advice everywhere this issue has surfaced: never cite a machine-generated case without independently confirming it against an official database such as the Supreme Court's own website, Indian Kanoon, or SCC Online; keep a research log capable of proving due diligence if the citation is ever questioned; and treat any such tool as a starting point for research, never as the final word on whether a precedent is real. What is learned as convenience at the start of a legal career can become genuine professional risk later. Law schools, in India and elsewhere, will have to adjust their approach, not by resisting the technology, but by teaching its limits more deliberately.

The Standard That Doesn't Change

In the end, the responsibility remains where it has always been. Lawyers are officers of the court, and that role carries a continuing duty of accuracy that predates any such technology by centuries. The ease with which information can now be generated does not reduce that obligation; if anything, it raises the bar. And for judges, the standard is identical: what enters judicial reasoning must be verified, even when it arrives with the appearance of unimpeachable authority.

AI will continue to become part of legal practice in India and everywhere else; that much is now beyond dispute, borne out by a Supreme Court judgment, a pending regulatory framework, a High Court circular, and a global tracker already past 1,400 documented failures. What remains uncertain is how responsibly the profession will use it. Because in law, credibility has never been created by speed, fluency, or confidence of language. It is created by proof. Verification was always part of justice. After July 2, 2026, no court, lawyer, or law student in India can credibly claim not to have been told so.