AI Made Up a Judgment. A Tribunal Cited It. The Supreme Court Said: That Is No Decision at All.

What happened when NCLT and NCLAT relied on fake AI-generated precedents in an insolvency case, and what India's highest court just declared about AI's role in legal proceedings

A Judgment That Should Not Exist, Cited in a Judgment That Got Set Aside

On July 2, 2026, the Supreme Court of India set aside two orders of the National Company Law Tribunal and the National Company Law Appellate Tribunal in an insolvency case. Not because the law was wrong. Not because the facts were disputed. But because both tribunals had relied on judicial precedents that do not exist.

The precedents were fake. They were hallucinated by an AI tool, presented as real case law, cited in tribunal orders, and used as the basis for legal reasoning that affected the rights of real parties in a live insolvency proceeding.

The Supreme Court's response was unambiguous. A decision based on non-existent, fake, or hallucinated material is no decision in the eyes of law. It declared zero tolerance for reliance on AI-generated precedents without verification, set aside both orders, and directed the Bar Council of India to constitute a committee to frame guiding principles and prescribe disciplinary consequences for advocates found submitting such material to courts.

This judgment, Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., is one of the most important rulings on AI and the legal profession to come from any court anywhere in the world.

What Actually Happened in This Case

The underlying dispute involves Essel Infraprojects Ltd., the corporate guarantor of Pan India Utilities Distribution Company Ltd. When PIUDCL defaulted on loan facilities from Jammu and Kashmir Bank, the Bank invoked the corporate guarantee and filed a Section 7 application under the Insolvency and Bankruptcy Code, 2016 before the NCLT Mumbai.

The NCLT admitted the application in August 2024. The suspended director of EIL challenged this before the NCLAT, arguing that EIL's liabilities had been transferred through demerger and amalgamation within the Essel group, and that a renewed sanction letter did not reference the corporate guarantee, implying its discharge. The NCLAT rejected these arguments and dismissed the appeal in September 2025.

The case reached the Supreme Court. And there, something unusual emerged: both the NCLT and NCLAT had cited judicial precedents in support of their reasoning that, when checked, did not exist. These were AI-generated hallucinations, produced by a language model and presented as real case law, either by counsel before the tribunals or embedded in the tribunal's own research process.

The Supreme Court found this sufficient to vitiate the entire adjudicatory process, regardless of whether those fake citations had directly determined the outcome.

What Is an AI Hallucination and Why Is It So Dangerous in Court?

An AI hallucination is when an AI language model generates information that sounds plausible and well-formed but is factually wrong or entirely fabricated. In everyday use, this might mean an AI incorrectly stating a historical date or describing a product feature that does not exist. In a legal context, it means an AI generating a case name, citation, judge's name, and summary of a judgment that never happened.

This is not a bug that has been fixed. It is an inherent feature of how large language models work. They predict the next most plausible word or phrase based on patterns in training data. When asked about a specific legal precedent they have not been trained on, they do not say "I don't know." They generate a plausible-sounding answer.

For a lawyer under deadline pressure, an AI-generated case summary can look completely real. It has the right format: a case name, a year, a court, a paragraph of reasoning. Checking whether the case actually exists requires going to a verified legal database, searching for the exact citation, and confirming that the judgment text matches what the AI described. Many lawyers skip this step.

When that happens in a courtroom or tribunal, the consequences are exactly what this case illustrates: a decision is made on the basis of precedent that does not exist, and the decision is no decision in the eyes of law.

What the Supreme Court Said: Four Key Holdings

First: Zero tolerance, no exceptions.

The Court declared that courts and tribunals must operate in "zero-tolerance mode" for producing, citing, or relying upon AI-generated precedents without verification. The word "zero" is doing real work here. The Court did not say "limited tolerance" or "case-by-case assessment." It said zero. No fake citation, regardless of its role in the decision, is acceptable.

Second: Even an iota of fake material vitiates the decision.

This is the most striking legal principle in the judgment. The Court held that even a single piece of non-existent or hallucinated material entering the decision-making process vitiates the adjudication as a whole, whether or not that material was decisive. The fake precedent does not have to be the reason the tribunal ruled the way it did. Its presence alone is enough to render the decision void in law.

This is an extremely high standard. It means that a tribunal which cites five real cases and one hallucinated case has produced a decision that is no decision in law, even if the real cases fully support the outcome.

Third: AI may be used as an aid, but human control is mandatory.

The Court was careful not to ban AI from legal practice. It explicitly clarified that its judgment does not restrict the rightful use of artificial intelligence as an aid. But the word "aid" is critical. AI can assist lawyers and judges in research, drafting, and analysis. What it cannot do is replace human verification of the material it produces. Judicial decision-making must remain under human control at every step.

Fourth: The Bar Council of India must act.

The Court directed the BCI to constitute a committee to examine the issue of fake AI-generated citations in Indian courts, frame guiding principles for the responsible use of AI in legal practice, and prescribe disciplinary consequences for advocates who present such material before courts.

This is the Court treating the problem not as an isolated incident but as a systemic risk requiring institutional response. A single judgment setting aside one order is a correction. Directing the BCI to create enforceable disciplinary standards is a framework.

This Is Not the First Time

The Supreme Court's observation, "this is yet again a case," signals that this is not an isolated incident. In March 2026, the Court had already raised concerns about AI-generated fake precedents appearing in proceedings. The language in yesterday's judgment, "yet again," tells us that the warning in March was not heeded by all.

This pattern matches what has happened globally. In the United States, attorney Steven Schwartz was sanctioned in 2023 for submitting AI-generated fake case citations in a New York federal court. In Australia, a barrister received a formal reprimand for citing non-existent cases. In the UK, the courts have issued practice guidance specifically warning against unverified AI-generated citations. In Canada, a judge publicly rebuked a lawyer for the same conduct.

What is different about the Indian Supreme Court's judgment is its breadth. The US and Australian cases resulted in individual sanctions. This judgment creates a legal principle, that AI-hallucinated citations render a decision void, that now applies across every court and tribunal in India.

What This Means for Every Lawyer Using AI

The message for practising lawyers is direct and non-negotiable.

Every case citation generated or suggested by an AI tool must be independently verified against a authorised and verified legal database before it is placed before any court or tribunal. SCC Online, Manupatra, or the official judgment repositories of the Supreme Court and High Courts are the standard verification sources. Verification means confirming that the case exists, that the citation is accurate, and that the legal proposition attributed to it appears in the actual judgment text.

Using AI to draft arguments is fine. Using AI to suggest relevant cases is fine. Placing those cases before a court without verification is not fine. It is, after this judgment, a potential ground for disciplinary action under BCI rules and a cause for the proceedings in which you appear to be declared void.

For law firms, this is an immediate compliance obligation. If your firm uses AI-assisted legal research, the workflow must include a mandatory human verification step for every citation before any document is filed. That verification step should be documented. If a case is challenged, you need to be able to show that you verified it.

What This Means for Courts and Tribunals

The Court's direction extends beyond lawyers. The NCLT and NCLAT in this case were themselves relying on non-existent citations. Whether those citations were introduced by counsel or sourced through the tribunal's own research process, the tribunals bear responsibility for the material in their orders.

Going forward, judges and presiding officers must apply the same verification standard to any case they cite in their own reasoning. A judgment that includes AI-generated hallucinations in its own analysis, rather than in citations submitted by parties, is equally void under the principle this Court has articulated.

The Bigger Question This Judgment Raises

The Supreme Court has drawn a clear line: AI is a tool, not a decision-maker. The human lawyer, the human judge, and the human adjudicator remain responsible for every citation, every proposition of law, and every piece of reasoning in every document before a court.

This makes sense as a legal principle. The entire structure of legal precedent depends on the reliability of citation. If citations can be fabricated by machines and inserted into judicial records without detection, the chain of precedent, the system through which law builds on itself over time, breaks down. The law is only as reliable as the citations on which it is built.

What this judgment does not fully resolve is the question of how courts will detect AI hallucinations before, rather than after, they affect proceedings. The answer, for now, is individual verification by responsible lawyers. But as AI tools become more embedded in legal practice, the institutional infrastructure for verification will need to catch up.

The Bar Council of India committee that this Court has directed will be one part of that infrastructure. What it recommends, and whether the recommendations have real teeth, will shape how Indian legal practice navigates AI for years to come.

The Bottom Line

A tribunal cited a case that does not exist. The Supreme Court found out. The orders are gone. The matter goes back to square one. And every lawyer in India now knows that presenting an unverified AI-generated citation to a court is not just an ethical lapse. It is grounds for the entire proceedings to be set aside and for personal disciplinary consequences under Bar Council rules.

Verify everything. Every time.

This Blog is for general informational purposes and does not constitute legal advice. For guidance on legal practice compliance, AI-related legal risks, or court proceedings, please contact our team.