AI-Hallucinated Precedents And The Integrity Of Adjudication In India

Introduction

Artificial intelligence is increasingly being used for legal research, drafting, translation and document review. These tools can improve efficiency, but they also produce a serious risk in adjudication: the generation of non-existent or inaccurately described case law. A fabricated judgment may appear authentic because it contains a plausible case name, citation, court and legal proposition. If it is not verified and is relied upon by a court or tribunal, the problem is no longer merely technological. It becomes a threat to the legitimacy of judicial decision-making.

The Supreme Court’s decision in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd provides the clearest warning yet from an Indian court. On 2 July 2026, the Court set aside orders of the National Company Law Tribunal and the National Company Law Appellate Tribunal after finding that the adjudicatory reasoning relied upon fake, non-existent and AI-hallucinated precedents.

The Court held that a decision based on such material is no decision in the eyes of law and amounts to a subversion of the rule of law. The judgment does not prohibit the use of AI in legal work. Its principle is more precise: AI may assist research and adjudication, but no advocate, judge or tribunal may present or rely upon unverified machine-generated material as authentic legal authority.

What AI Hallucination Means in Legal Research

Generative AI systems predict language based on patterns in their training data. They do not necessarily retrieve and verify judgments from an authoritative database. When asked to provide authorities on a legal issue, a system may generate a convincing but non-existent case, combine facts from different decisions or attribute a genuine proposition to the wrong authority.

In ordinary writing, an inaccurate citation may be corrected later. In adjudication, the consequences are more serious because precedent performs a constitutional and institutional function. Courts do not merely cite cases as decoration. They use judicial decisions to identify legal rules, maintain consistency, explain the basis of conclusions and demonstrate that similar cases are treated alike.

A fabricated precedent can therefore distort the legal test applied to a dispute. It may create an imaginary exception, reverse the meaning of a statute or give an appearance of authority to a conclusion that has no legal foundation. The danger increases where the opposing party does not have an adequate opportunity to verify the material or where the tribunal adopts it without independent reasoning.

The Pooja Ramesh Singh Decision

The dispute arose in insolvency proceedings involving Jammu and Kashmir Bank. The NCLT relied on several authorities in reaching its decision. The Supreme Court found that some cited judgments did not exist, while others either did not contain the legal propositions attributed to them or were irrelevant to the context in which they were cited.

The Court described this as reliance on material generated through AI and presented as precedent. It set aside the NCLT order and the NCLAT judgment and restored the matter for fresh consideration on the merits. The result was not based on a minor citation error. The decision was invalidated because the adjudicatory process had been influenced by material that had no legal existence.

The Court emphasised that the defect struck at the integrity of adjudication. A judicial decision must be founded on real facts, applicable law and authentic authorities. If the reasoning depends, even in part, on fabricated precedent, the court cannot assume that the conclusion would have been the same without it.

The Court also directed attention to professional responsibility. The use of AI is not itself misconduct, but citing generated material without verification is a serious lapse. Advocates have a duty of candour to the court and must ensure that authorities submitted as precedent are genuine and accurately represented.

Why Verification Is Essential

Precedent derives its authority from an actual judicial act. A real judgment records the parties, facts, submissions, reasoning and result. It can be examined, distinguished, followed or overruled. A fabricated case has none of these characteristics. It cannot contribute to the development of law because it has no institutional origin.

Verification is therefore not a technical formality. It is the step that separates legal research from plausible text generation. A citation must be checked against an official court website, a recognised legal database or a reliable report. The researcher must read the judgment itself and confirm that the quoted proposition appears in the relevant paragraph and applies to the factual context.

The problem is not limited to entirely invented cases. A genuine decision may be misquoted, taken out of context or cited for a proposition it never decided. Such misuse can be equally damaging because it gives false authority to an argument while making detection more difficult.

The Rule of Law Consequences

The rule of law requires that legal outcomes be based on publicly ascertainable norms rather than personal preference or undisclosed material. Authentic precedent supports this requirement by making judicial reasoning transparent and reviewable.

AI-hallucinated authorities undermine each element. They make the law appear more certain than it is, prevent meaningful appellate review and create unequal advantages for parties able to introduce unverified material. If one party’s fabricated citation is accepted while another party relies on authentic law, the fairness of the proceeding is compromised.

The problem also affects judicial independence. Courts must decide according to law, not according to the apparent confidence of a generated text. A citation that looks authoritative may influence the decision-maker even if its underlying reasoning is absent. Human judgment remains essential because only a judge can determine whether an authority is genuine, binding, distinguishable or persuasive.

The Supreme Court’s language that such a decision is no decision in the eyes of law reflects this institutional concern. The defect is not simply that the tribunal reached a wrong conclusion. The adjudication was conducted through an unreliable legal foundation.

Professional Responsibility of Advocates

Advocates are responsible for the accuracy of the authorities they cite. Using AI may be compatible with professional practice, but responsibility cannot be delegated to the software. An advocate must verify every case name, citation, quotation, paragraph reference and legal proposition before filing.

A proper internal process should include source verification by a second lawyer, particularly for authorities generated through AI-assisted research. Citations should be checked against official judgments, and the final pleading should contain only authorities that counsel has read.

The use of fabricated authorities may expose an advocate to adverse orders, disciplinary proceedings, costs, professional reputational damage and, in serious cases, contempt-related consequences. The Supreme Court’s direction to the Bar Council of India to deliberate on the consequences of submitting fake or hallucinated precedents underscores that the issue concerns professional ethics, not merely drafting quality.

Responsibility of Courts and Tribunals

Judicial verification is also necessary. Courts and tribunals cannot assume that material placed before them is accurate, especially where the citation is unfamiliar or the proposition appears inconsistent with established law.

A judge is not expected to independently research every authority cited by counsel. However, where a decision relies materially on a case, the court should verify the authority, examine the relevant passages and ensure that it is applicable. The obligation becomes stronger where the judgment is obscure, the citation is incomplete or the legal proposition appears unusually broad.

Tribunals should also adopt institutional protocols for AI-assisted research. Judges and members may use digital tools to locate authorities, but the final decision should be based on verified primary sources. A citation-management system that links each authority to the official judgment can reduce error.

Other Relevant Precedents

Indian courts have previously shown that procedural and adjudicatory decisions must be based on authentic legal material. In State of Punjab v. Davinder Pal Singh Bhullar, the Supreme Court emphasised the importance of judicial reasoning and the consequences of decisions rendered without proper legal consideration. While the case did not concern AI, its broader principle supports the requirement that adjudication be reasoned and legally grounded.

In Supreme Court Bar Association v. Union of India, the Court distinguished judicial power from professional disciplinary power and recognised the importance of maintaining institutional standards in the administration of justice. The decision is relevant because AI-related citation failures may require both correction of the judgment and consideration of professional consequences.

The broader comparative experience also supports caution. Courts in the United States and the United Kingdom have issued warnings and procedural directions after lawyers cited fabricated AI-generated cases. These developments reinforce that the problem is global, but the solution remains consistent: human verification, disclosure where required and responsibility for the accuracy of submissions.

Judicial Use of AI

The Supreme Court’s judgment should not be read as opposition to legal technology. AI can assist with translation, document classification, chronology, retrieval and identification of potentially relevant authorities. It may help judges and lawyers manage large records and improve access to legal information.

The danger arises when assistance becomes substitution. AI cannot determine whether a case is binding, whether a factual distinction is material or whether a legal proposition remains good law. It may also fail to identify whether a judgment has been overruled, stayed or confined to its facts.

A responsible system should therefore use AI in a controlled manner. The user should treat generated authorities as leads, not sources. Every authority must be verified through an authoritative database. Generated summaries should be compared with the judgment. Sensitive case materials should not be uploaded to systems that lack appropriate confidentiality and security protections.

Consequences for Judicial Decision-Making

The immediate consequence of relying on fabricated precedent is that the affected decision may be set aside. The court may remand the matter for fresh consideration, as occurred in Pooja Ramesh Singh. This causes delay, additional costs and uncertainty for the parties.

The institutional consequences are broader. Repeated reliance on false authorities can reduce public confidence, weaken the authority of tribunals and create doubts about the reliability of published judgments. It may also burden appellate courts with proceedings that would have been unnecessary if the original citations had been verified.

The judgment therefore establishes a preventive principle. Accuracy in legal citation is a condition of legitimate adjudication. Courts and counsel must ensure that every authority relied upon exists, says what it is claimed to say and is relevant to the dispute.

Conclusion

Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd establishes a strict judicial standard for AI-assisted legal research. A court or tribunal cannot rely on fake, non-existent or hallucinated precedents and still treat the resulting decision as legally valid. The Supreme Court’s zero-tolerance approach protects the authenticity of precedent, the fairness of proceedings and the rule of law.

The judgment does not ban AI. It requires human control and verification. Advocates remain responsible for the authorities they cite, while judges and tribunals must ensure that the foundations of their decisions are real and legally relevant.

AI may improve the speed of legal research, but it cannot replace professional judgment. The essential rule is simple: generated legal material may guide inquiry, but only verified judicial authority may support adjudication.

References

1. Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., 2026 INSC 668.
2. State of Punjab v. Davinder Pal Singh Bhullar, (2011) 14 SCC 770.
3. Supreme Court Bar Association v. Union of India, (1998) 4 SCC 409.
4. Union of India v. R Gandhi, President, Madras Bar Association, (2010) 11 SCC 1.
5. Assistant Commissioner, Commercial Tax Department v. Shukla & Brothers, (2010) 4 SCC 785.
6. Kranti Associates Pvt Ltd v. Masood Ahmed Khan, (2010) 9 SCC 496.
7. Bar Council of India Rules, Part VI, Chapter II, Standards of Professional Conduct and Etiquette.
8. Constitution of India, Article 14 and 21.
9. Maneka Gandhi v. Union of India, (1978) 1 SCC 248.
10. Law Society of England and Wales, Generative AI: The Essential Guide for Lawyers (2025).

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.