ACCOUNTABILITY BETWEEN BAR AND BENCH FOR AI-HALLUCINATED PRECEDENTS

 

On 2 July 2026, in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. (2026 INSC 668), the Supreme Court set aside the orders of the NCLT and NCLAT admitting a Section 7 insolvency application against Essel Infraprojects Ltd., a corporate guarantor for the defaulted loans of Pan India Utilities Distribution Company Ltd. The NCLT had admitted the application, and the NCLAT had upheld that decision. Both relied on six cited precedents.

When the Supreme Court independently examined those citations, however, it found something deeply troubling. Three of the six precedents did not exist at all. Two were genuine judgments, but the paragraphs attributed to them could not be found anywhere in the reported decisions. The sixth citation referred to a real judgment, but to an entirely different case. The Supreme Court consequently remanded the matter to the NCLT for fresh disposal, while making it clear that it was expressing no opinion on the merits.

The problem here goes beyond an error in citation. Judgments built on non-existent or misrepresented case law undermine confidence in the judicial process. They also have tangible consequences. When proceedings have to be reopened because the law relied upon was never law at all, litigants lose time, incur additional costs and, in some cases, remain caught in litigation that should never have taken the course it did.

The Supreme Court has rightly treated the use of fabricated precedents as a serious matter. But the question is whether the response addresses the entire problem.

The Court's directions place the primary burden on the Bar. An advocate who cites material that does not exist may face disciplinary consequences, and the Bar Council of India has been directed to examine the conduct of the advocates involved. What remains less clear is the corresponding accountability on the judicial side. The judgment describes the reliance upon such material by the tribunal as a "serious lapse", but does not establish a comparable mechanism for examining how the material entered the tribunal's own research process or why it escaped scrutiny at the adjudicatory and appellate levels.

This distinction matters because, according to the judgment itself, the problem did not necessarily originate with the advocates.

Paragraph 16 records that the precedents were "obtained through its research" and were not submitted by the advocates. The Court further observed that the fake judgments had "escaped scrutiny by the statutory appellate tribunal." In other words, the material originated in the tribunal's own research process and then passed through two levels of judicial scrutiny without being detected.

That makes the issue considerably broader than professional misconduct at the Bar. It raises a question about how courts and tribunals themselves research, verify and incorporate authorities into their judgments in an age where AI-generated legal content can look remarkably convincing while being entirely fictitious.

The judgment's opening paragraph makes this concern even more significant. The Court refers to the tribunal relying upon such material "yet again". That single word is important. It suggests that the Court is not confronting an isolated incident, but a recurring problem. If fabricated authorities are repeatedly entering judicial reasoning, the solution cannot rest solely on disciplining the person who placed a citation before the court. The judicial process through which that material was researched, accepted and ultimately reproduced must also be examined.

One could argue that the apparent asymmetry is structural. Judicial accountability operates through different institutional channels, including mechanisms such as contempt jurisdiction and administrative supervision. A civil appeal may not be the appropriate forum to impose disciplinary consequences upon a judge or tribunal member. That may explain why the Court did not proceed further against the judicial officers concerned.

But it does not answer the institutional question.

Having identified the tribunal's reliance on fabricated material as a "serious lapse", the Court could also have addressed the safeguards required to prevent such material from entering judgments in the first place. It identified the problem on the Bench, but directed concrete disciplinary scrutiny primarily towards the Bar. That is not merely a question of jurisdiction. It is also a question of where responsibility for verification is placed.

A useful comparison can be found in England and Wales, where judicial guidance on the use of AI was updated in October 2025. Importantly, that guidance is not directed only at advocates. It expressly addresses judges and members of the judicial team, including clerks, judicial assistants, legal advisers and other support staff. The underlying approach recognises that AI-generated material creates a risk across the judicial process, not merely within legal submissions made by counsel.

That approach offers a useful lesson. If AI can generate a convincing but non-existent judgment, verification cannot be treated as the exclusive responsibility of the person who happens to cite it. Verification has to be built into the system at every stage at which legal authorities are researched, relied upon or incorporated into a judgment.

Pooja Ramesh Singh therefore presents an important opportunity to reconsider how courts and tribunals deal with AI-generated legal research.

The direction to the Bar Council in paragraph 9 is necessary. Lawyers cannot be permitted to place fictitious authorities before a court and then escape responsibility simply because the material was generated by an AI tool. Professional responsibility remains essential.

But it cannot be the entire answer.

The Supreme Court, or the High Courts while exercising their supervisory jurisdiction, should consider requiring tribunals and appellate bodies to introduce a basic verification function within their existing judicial infrastructure. Law clerks, research assistants, legal advisers or another designated member of the judicial team could be tasked with verifying cited authorities before they become part of a final judgment. Such a mechanism would not replace the judge's responsibility; it would strengthen it.

The objective should not be to create another layer of bureaucracy. It should be to ensure that a case which does not exist never becomes a precedent in a judgment.

This is particularly important because fabricated authorities can survive surprisingly far into the judicial process. In Pooja Ramesh Singh, the material escaped scrutiny at the tribunal stage and again at the appellate stage. If two levels of adjudication can unknowingly rely upon non-existent judgments, simply asking lawyers to exercise greater caution will not eliminate the systemic risk.

The consequences of getting this wrong are borne by litigants. They are the ones who wait while proceedings are reopened. They bear the additional legal costs. They experience uncertainty created by decisions resting on authorities that were never part of the law at all.

AI has changed the speed and scale at which legal research can be conducted. It has also changed the nature of the verification problem. A fabricated judgment generated in seconds can look sufficiently authentic to survive an initial reading, particularly when it contains a plausible case name, citation, court, date and paragraph number. That is precisely why the response must be institutional rather than merely disciplinary.

The judiciary has declared a need for "zero tolerance" towards fabricated precedents. That principle is welcome. But zero tolerance must operate across the courtroom.

The Bar must verify what it puts before the court. The Bench must verify what it relies upon. And the judicial system must have safeguards capable of catching errors before they become part of a judgment.

Otherwise, the risk is that the next fabricated precedent will not come from an advocate at all. It may emerge from the research process of the court itself.

And when that happens, there may be no advocate left to blame.

Until the Bench is held to the same standard of verification that it expects from the Bar, the "zero tolerance" declared by the Court will govern only half the courtroom. 

 Share