Fabricated Precedents, Real Consequences: Indian Judiciary Confronts the Generative AI Problem

ABSTRACT

Recently, the Indian judiciary faced a problem: the developmental gap between adjudicatory and institutional bodies. Litigants are relying on AI-generated content, ignoring the inherent caution of hallucinated texts. The article focuses on two main Supreme Court judgments of 2026, Gummadi Usha Rani v. Sure Mallikarjuna Rao and Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., noting the court’s intent and critical shift in determining liability for using “hallucinated “ citations. These cases denote an application of strict liability & voidability of fabricated materials, treating them as “disruption of the rule of law” rather than clerical error. The article attempts to analyze the ‘dual-track accountability framework’ created by the Supreme Court and imposed on institutional bodies. Comparing the Indian approach with the United Kingdom’s calibrated approach, which accounts for the bad intent carried behind, unlike the former’s stricter approach, which imposes liability due to the mere presence of fabricated material. The article draws upon Rebecca Crootof’s theory on “technological-legal lock-in,” arguing the court’s direction as proactive conduct seeking to prevent institutional dependence on AI fabrications.Finally, the article criticizes the capacity to carry out the zero-tolerance doctrine throughout the adjudicatory levels. The successful implementation of the doctrine depends upon the formulation of verification portals by the Bar Council of India, and in the absence of clear distribution channels, there may be a huge financial burden. This doctrinal transition of liability marks the judiciary’s intent to treat AI errors seriously affecting the ‘sanctity of adjudication’ instead of viewing them as an ordinary diversion.

INTRODUCTION

The Supreme Court of India was confronted with the same institutional embarrassment twice between February and July 2026 – Judgments relied on non-existent case laws. Earlier this year, a trial court in Andhra Pradesh dismissed the objections to an Advocate Commissioner’s report that contained four fabricated case laws. The High Court had cautioned officers of law while using artificially generated sources and to proceed with cases on merits. Later in July, the Court found that the National Company Law Tribunal & the National Company Law Appellate Tribunal had supported an insolvency order under Section 7 of the Insolvency and Bankruptcy Code, 2016 on multiple citations, which were false and defective in the case of Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., 2026. These failures reflect poor standards of care and caution in using generative AI in litigation. Its tendency to hallucinate has moved from a law-school warning statement into live adjudication at a tribunal level dealing with real-life commercial matters. This article examines the doctrinal response fashioned by the Supreme Court across these matters and asks whether the remedy is administrable at scale.

LEGAL ANALYSIS

The Bench of Narasimha and Aradhe, JJ., in the Pooja Ramesh Singh issue, articulated the effect of AI-fabricated authority: a decision “based on material which is fake and hallucinated is no decision at all,” amounting to “subversion of the rule of law,” and liable to be set aside “even if an iota” of such material enters, irrespective of its nexus on the outcome. The tone of the court represents the gravity and tolerance to the use of AI, regardless of its contribution to the outcome. In addition, the court reiterated the point using an unusual metaphor – “invisible, insidious and catastrophic by the time anyone notices”.

Dual-Track Accountability Framework
Accountability under this judgment runs on two tracks. Firstly, holding advocates liable for misconduct rather than mere oversight, while citing AI content. Secondly, the significant liability on a judge or tribunal that relies on such fabrication commits a ‘serious lapse’. The Court’s refusal to position the standards at the Bar, where negligence is prone, and instead places it for the adjudicators, who are the last line of verification, depicts zero tolerance for AI citations. In the Pooja Ramesh Singh case, the court directed the Bar Council of India to constitute a committee to raise awareness on hallucinated-citation practices among advocates and to formulate disciplinary plans. Before this case, the court, through the Gummadi Usha Rani proceedings, issued a notice to the Advocate General, Solicitor General, and the Bar Council, and appointed senior counsel to aid the court in this matter. Such a measure is generally reserved for policy-level interventions for the Court, rather than case-specific relief.

This conduct was seen as worthy of recognition, since the exercise of legal analysis is not merely descriptive. A norm that nullifies the decision for the mere presence of fabrication without inquiry is doctrinally precise but administratively expensive. It lures every litigant who has lost before the court to scrutinize the cited sources for defects, regardless of their contribution to the reasoning. The Court itself acknowledged that “mere declaration of prohibitory action is not sufficient” and that consequential action must follow. But, in the case of Pooja Ramesh Singh, the underlying failure was the absence of direction to one who bears the verification task. Neither has the Bar Council stated a timeline, and until it reports, the zero-tolerance rule operates without a complementary detection system, prone to the problem that the Courts warned against.

Looking at England’s approach to the tussle between AI and the rule of law, they have adopted disciplinary actions modeled on the intent carried. A purely AI-driven law firm in the United Kingdom was under the regulation that specifically required the firm to manage ‘the risk of AI hallucinations’ before authorizing, and drafted a framework of continuous supervision & accountability for AI output. For example, a presiding judge declined to initiate contempt proceedings for want of dishonest intent, but issued a public reprimand. Such an approach displays orders calibrated to intent, unlike the Indian approach that relies on strict liability, voiding the decision regardless of materiality. The Supreme Court failed, except for its general concern for the ‘sanctity of adjudication,’ to explain why such an approach was adopted.

The confrontation between generative AI and the courts of law was addressed well before such a situation in the Indian courts. Once an adjudicating system merges with an institutional & infrastructural mechanism that depends on technology, the correcting course becomes progressively harder despite the knowledge of the flaws, because such a reversal carries its financial burden. The adjudication can be seen as a proactive strike to prevent further dependence on AI outputs that are unverified before utilizing them in the court of law.

SUPPORTING AUTHORITY

The article doctrinally rests upon two decisions of the Supreme Court, delivered in the span of 6 months by the same coram, as a two-part statement rather than a single ruling. The first arose from a civil revision by the High Court that discovered fabricated citations that reached the subordinate courts. The second stems from an insolvency proceeding, giving the court the first real opportunity to use it as an occasion to treat the problem, preserving adjudicatory integrity. The common nexus is the involvement of executive bodies: the courts directed the Bar Council committee to act as a regulatory body on the issue and placed the Union’s lawyers as senior counsel on notice.

The comparative context of a foreign court’s response and Indian approach reflects the foundational intent of not tolerating AI hallucinations that impede adjudications. The measurements display a disciplinary model calibrated to intent, one that stands just short of automatic nullification and the other that refuses to proceed in the mere presence of fabrication regardless of its contribution to the reasoning.

CONCLUSION

Both the cases decided within 6 months of each other establish the Indian Judiciary’s intent to treat AI hallucinations as a matter of adjudicatory integrity rather than clerical error, subsequently voiding fabricated material, regardless of its influence on the outcome. The Supreme Court has taken a stricter approach to disciplinary action when compared to the survey in the United Kingdom, but ironically the successful implementation depends upon what the Bar Council committee produces and on whether the court, on account of “Public Policy and enforceable Rules and Regulations,” will create a verification body at problem sources. Until such an action, parties in question have an incentive to audit citations that were relied upon, which is a form of self-help that may act as immediate relief when compared to the zero-tolerance doctrine.