India's Supreme Court has set aside lower court orders built on invented precedents and asked the country's lawyer licensing body to treat reliance on AI fabricated case law as professional misconduct.
India's Supreme Court has caught something that should not have been possible. On July 2, 2026, Justices P.S. Narasimha and Alok Aradhe set aside orders from the National Company Law Tribunal (NCLT) and the National Company Law Appellate Tribunal (NCLAT) in the Essel Infraprojects insolvency matter because six of the precedents the tribunal had relied on were either invented or corrupted. Three of the citations did not exist at all. Three more were real judgments carrying paragraphs and case titles that no one had written.
The bench's description of what had happened was unusually blunt. Hallucinated citations are a "catastrophic" and "invisible" failure of the judicial process, the Court said, likening them to "the release of methyl isocyanate in the province of law and justice," the gas that killed thousands in Bhopal in 1984. The Court treated reliance on AI-fabricated case law not as error but as misconduct with legal consequences for advocates, and asked the Bar Council of India (BCI), the country's national lawyer-licensing body, to frame disciplinary norms for the practice. (The Hindu, Hindustan Times)
The Essel matter is the most instructive case in the cluster because the contamination did not arrive in a lawyer's brief. According to an affidavit from Jammu and Kashmir Bank, its counsel had cited none of the six judgments the NCLT relied on. The tribunal sourced the citations through its own research, applied them as if they were settled precedent, and produced an order that became enforceable against the parties. The NCLAT reviewed the order on appeal and missed all six. Only the Supreme Court caught them. (India Today, LiveLaw)
That detail changes the shape of the problem. Most professional-conduct frameworks, including the disciplinary norms the BCI may now draft, target what lawyers file. The Essel facts show that the same failure can occur inside the court's own research process, with no advocate in the loop. Existing misconduct rules do not cover that vector.
The Essel order is the Supreme Court's first zero-tolerance disposition on the issue, but it is not the Court's first encounter. In February 2026, a bench led by Chief Justice Surya Kant flagged the non-existent case "Mercy v. Mankind" while hearing a PIL on political speeches; Justice B.V. Nagarathna said the citation had surfaced in her court, and the Chief Justice noted a "series of such judgments" had appeared before Justice Dipankar Datta. A week later, the Court took cognisance of a Gummadi Usha Rani trial-court order based on fake AI-generated judgments, stayed reliance on the disputed report, and appointed senior advocate Shyam Divan as amicus curiae. In a separate matter, the Bombay High Court fined a litigant Rs 50,000 for citing fabricated AI-generated judgments. (Hindustan Times, LiveLaw, Medianama)
The Court's reference to a national rule is not a request from a blank slate. The Court had previously circulated draft rules on AI use in the legal process; the July 2 ruling is the first time it has ordered a statutory body to act on them. The framing of the issue as misconduct rather than error is the more consequential lever, because it pulls AI misuse inside existing advocate-discipline machinery without waiting for new legislation.
What remains unresolved is exactly the failure mode the Essel case exposes: AI-generated research tools used inside tribunals and trial courts, where no professional-conduct framework applies and where the contamination enters the record without a human advocate having ever cited the fabricated precedent. The BCI referral covers what lawyers file. It does not yet cover what courts read.
The next trigger to watch is the Bar Council of India's response to the Court's request. A rule that disciplines advocates for unverified AI citations would close one half of the vector. Closing the other half, the court's own research process, is a harder institutional problem with no disciplinary handle yet in sight.