234 incidents on record · 2026HeadlightsIncident reports by Ellie Harris · Melbourne
5 new this weekLibrary last updated 14 October 2026
← The incident library
HD-INC-207
Courts and justice · India · 2026 · An adjudicating tribunal relied on non-existent and misquoted precedents, described by the Supreme Court as AI-generated, and the appeal tribunal repeated them without detecting the problem

India's Supreme Court Voided a Tribunal Order Built on Fake AI-Generated Precedents

By Ellie Harris · Filed Tribunal order of 28 August 2024; appeal tribunal order of 11 September 2025

Alleged: National Company Law Tribunal, Mumbai; National Company Law Appellate Tribunal developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

India's Supreme Court Voided a Tribunal Order Built on Fake AI-Generated Precedents

What happened

The Supreme Court of India’s judgment of 2 July 2026 records that a company had guaranteed a bank loan to another company, which failed to repay it. The National Company Law Tribunal in Mumbai admitted the guarantor into insolvency proceedings on 28 August 2024.

The judgment records that the tribunal relied on six authorities from earlier Supreme Court cases. When the Supreme Court examined them, three citations did not exist. Two referred to real cases but included paragraphs that could not be found in those judgments. One gave the wrong citation for a real case, again with a paragraph that does not exist.

The appeal tribunal upheld the decision on 11 September 2025. Its judgment repeated the same six authorities.

Before the Supreme Court, counsel for the appellant, a suspended director of the guarantor company, submitted that the authorities were fake and “probably AI-generated”. The bank filed an affidavit saying its lawyers had not cited them, and that they had come from the tribunal’s own research.

The Supreme Court set aside both decisions and sent the insolvency application back to the tribunal to be decided again. It did not decide whether the company should be placed into insolvency.

What an auditable version would have shown

The Supreme Court described the material as “non-existent, fake and hallucinated material, generated through Artificial Intelligence (AI)”. According to the judgment, some AI-generated passages were wrongly attributed to genuine citations.

The judgment does not identify which AI tool was used or who used it. The bank’s affidavit said the authorities came from the tribunal’s own research, but the court made no finding about who produced them.

A record of the tribunal’s drafting would have shown where each quoted passage came from, who checked it and whether the words appear in the original judgment.

The Supreme Court examined the authorities itself. It set out where each one appeared in the tribunal’s decision and what was wrong with it.

Where the gap was

According to the judgment, the fake judgments “escaped scrutiny by the first statutory appellate tribunal”. The court said that courts and tribunals generally trust lawyers when they cite precedents. Here, according to the bank’s affidavit, the lawyers had not cited them.

The judgment does not record any check of the six authorities against the law reports at either tribunal, and the same authorities travelled from the first decision into the appeal.

A CitationVerifier is designed to check whether a cited authority exists and whether the quoted text appears in it before a document is issued.

A ConductRecord keeps the source of each quoted passage, the check and the result.

CitationVerifier and ConductRecord are Headlights designs.

What governance should have looked like

The Supreme Court declared “zero tolerance” for citing or relying on unverified AI-generated precedents. It said a judge who relies on such material commits a serious lapse, and that a decision based on it “is no decision in the eyes of the law, irrespective of whether such material had a direct or indirect bearing on the decision-making”. It said its judgment has no bearing on the rightful use of AI.

It directed the Bar Council of India to set up a committee to prescribe guidance, and disciplinary consequences, for advocates who submit such material. It made no equivalent direction for tribunals.

For a court or tribunal using AI in research or drafting, best practice would be to check that every cited case exists and that every quotation matches the original judgment, and to record those checks before the decision is issued. An appeal tribunal should check the authorities in the decision under appeal, not only the reasoning.

Failure Pattern: an insolvency tribunal relied on non-existent and misquoted precedents that the Supreme Court described as AI-generated, and the appeal tribunal repeated them without detecting the problem.

Governance Principle: a court or tribunal should be able to show, for every authority it relies on, where the text came from and that someone checked it against the official report before the decision was issued.

The reference implementation of CitationVerifier and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

Sent through Substack. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and National Company Law Tribunal, Mumbai; National Company Law Appellate Tribunal could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

Last reviewed September 2026. This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →