New Delhi, July 2026. In a ruling that legal circles are already calling one of the sharpest judicial responses yet to the creeping problem of artificial intelligence hallucinations in Indian courtrooms, the Supreme Court has struck down orders passed by the National Company Law Tribunal (NCLT) and the National Company Law Appellate Tribunal (NCLAT) after discovering that the tribunals had leaned on court “precedents” that simply do not exist, fabricated wholesale by an AI tool and presented as genuine case law.

The judgment, delivered on July 2 by a bench of Justice Pamidighantam Sri Narasimha and Justice Alok Aradhe, arose out of an insolvency dispute: Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. But its real significance lies far beyond the specific facts of the case. The Court has now put every tribunal, every judge, and every advocate in the country on notice. Cite an AI-generated case that turns out to be fake, and the resulting order will not survive appellate scrutiny, no matter how minor the fabricated material’s role in the final decision.

How the Case Reached the Apex Court

The dispute traces back to a corporate guarantee. Pan India Utilities Distribution Company Ltd. (PIUDCL) had taken loan facilities from Jammu and Kashmir Bank, with Essel Infraprojects Ltd. (EIL) standing as the corporate guarantor. When PIUDCL’s finances deteriorated and its loan accounts turned into non-performing assets after repeated defaults, the bank moved under Section 7 of the Insolvency and Bankruptcy Code, seeking to bring EIL into insolvency proceedings as the guarantor liable for the debt.

The NCLT agreed with the bank. It found that debt and default had been established, admitted the insolvency application, appointed an Interim Resolution Professional, and imposed a moratorium under Section 14 of the Code, the standard sequence once a company is pushed into the corporate insolvency resolution process. EIL’s suspended director challenged that order before the NCLAT, the statutory appellate forum, but the appeal was dismissed, with the NCLAT effectively endorsing the NCLT’s reasoning, including its citation of prior “precedents.”

It was only when the matter reached the Supreme Court that the truth surfaced. Those precedents did not exist. It was AI-constructed garbage, invented names for cases and paragraphs that had been incorrectly mashed into real citations, and it had passed not only through the initial tribunal but through the appeal tribunal as well without anyone being tripped up.

What the Bench Found

According to the Court’s own account, the bank itself filed an affidavit clarifying that its lawyers had never cited the judgments the NCLT relied upon. In other words, the tribunal appears to have generated the bogus case law through what the Court described as its own independent research, raising the uncomfortable possibility that AI tools may have been used directly by the adjudicating body, not merely by counsel appearing before it.

The bench did not mince words about the scale of the problem this reveals. It noted that courts and tribunals currently operate on an implicit trust that lawyers are citing genuine precedent, and questioned how realistic it would be to expect judges to independently verify every single citation placed before them. But in this case, even the NCLAT, the check and balance in the system supposedly there to detect such a lapse, did not spot that the “precedents” on which the order being examined was based were completely fake.

The judges were unambiguous about the consequence: a decision built even partly on hallucinated material is not a decision that can stand. As the bench put it, such an outcome amounts to “subversion of the rule of law,” and would be set aside even if only a trace of fake or hallucinated material found its way into the reasoning. The ruling goes further than simply flagging an error. It frames the citation of an AI-produced and unchecked judgment explicitly as an offence of misconduct by the lawyer concerned, and as a dereliction of the most serious nature by any judge accepting it.

With that reasoning, the Court set aside both the NCLT’s original order and the NCLAT’s dismissal of the appeal, restoring the original Section 7 application to the NCLT’s docket for it to be decided afresh, this time, presumably, on real law.

Why This Ruling Matters Beyond One Insolvency Dispute

This is not the first time Indian courts have caught AI hallucinations slipping into judicial reasoning, but the framing here is unusually forceful. The Court called for what it termed a zero-tolerance approach across the judicial system, not a case-by-case tolerance for minor errors, but an institutional stance that fabricated precedent, however it enters the process, voids the outcome.

For litigants, the practical takeaway is significant. It can be overturned and go the other way with any AI-produced case law, even if only a very small proportion of the reasons are based on invented case law, and it makes very little or no difference to the actual outcome of the decision. For practitioners, the message is equally direct. Relying on an AI tool to locate case law without independently verifying that the judgment actually exists is no longer a grey area. It now carries the risk of being treated as misconduct.

The ruling also implicitly raises questions about the internal practices of tribunals like the NCLT and NCLAT, which handle a heavy volume of complex commercial and insolvency litigation. If research support, whether by law clerks, registry staff, or the tribunal members themselves, is being supplemented by generative AI tools without a verification layer, the same failure that occurred in this case could recur elsewhere, across any number of pending matters.

The Broader Pattern

This judgment lands amid a broader judicial reckoning with generative AI across the legal system. Courts in several jurisdictions have already sanctioned lawyers for filing briefs containing invented case citations, and Indian courts have increasingly flagged similar concerns in recent months. What sets this ruling apart is that the hallucinated material appears to have entered through the tribunal’s own research process rather than through a lawyer’s brief, shifting part of the scrutiny from the bar to the bench itself.

For law students and early-career practitioners tracking the intersection of AI and the legal profession, the case is likely to become a reference point for a fast-emerging body of jurisprudence on AI accountability in adjudication, one that will only grow as generative tools become more embedded in everyday legal research, drafting, and even judicial decision-making.

The Supreme Court has not banned the use of AI research tools in courts or tribunals. What it has done is make clear that convenience does not excuse verification, and that the burden of confirming a precedent, real or not, rests squarely on whoever cites it, lawyer or judge alike.

Cause Title: Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. (Neutral Citation: 2026 INSC 668), decided July 2, 2026, by a bench of Justice Pamidighantam Sri Narasimha and Justice Alok Aradhe.