The Judgment That Never Happened: A Note on the Supreme Court’s AI-Citation Ruling
Summary: Article discusses the Supreme Court’s decision in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. , where fabricated case citations, stated to have been generated through a tribunal’s own research, appeared in NCLT and NCLAT orders in an insolvency matter before being identified before the Supreme Court. The Court set aside the NCLT and NCLAT orders and remanded the Section 7 application for fresh disposal, stating that unverified AI-generated or fabricated case law cannot be relied upon, whether introduced by an advocate or through a court’s own research. According to the article, the Court held that advocates citing such material without verification commit professional misconduct and that reliance on fabricated material by a judge or tribunal member constitutes an equally serious lapse. The article further notes that the Court directed the Bar Council of India to constitute a committee to address advocates placing fake or hallucinated precedents before courts and prescribe disciplinary action. It also discusses the implications for AI-assisted legal research and suggests verification practices for citations generated by AI tools.
How a Fake Precedent Survived Two Tribunals
The underlying dispute was a fairly ordinary insolvency matter. Jammu and Kashmir Bank Ltd. had extended credit facilities to Pan India Utilities Distribution Company Ltd., secured by a corporate guarantee from Essel Infraprojects Ltd. Once the borrower’s account turned non-performing, the bank filed a Section 7 application against Essel Infraprojects as guarantor. The NCLT admitted it, appointed an interim resolution professional, and declared a moratorium in the usual way. Essel’s suspended director appealed to the NCLAT, which dismissed the appeal and endorsed the NCLT’s reasoning – including the precedents it had cited.
Here’s where it gets unusual. Before the Supreme Court, the bank itself filed an affidavit saying its own counsel had never cited the judgments in question. The precedents, it turned out, appeared to have come from the Tribunal’s own research. So the fabricated material hadn’t slipped in through an advocate cutting corners – it had entered through the adjudicating authority’s own process, which is arguably the harder problem to guard against.
What the Court Held
Justices P.S. Narasimha and Alok Aradhe, deciding Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. (2026 INSC 668) on 2 July 2026, set aside both the NCLT and NCLAT orders and sent the Section 7 application back for fresh disposal. But the reasoning reaches well past the four corners of this case. The Bench’s message to courts and tribunals was blunt: stop giving unverified AI-generated case law any benefit of the doubt, whether it comes from an advocate’s citation or from a tribunal’s own research. Producing it, relying on it, or letting it slip into an order unchecked all get treated the same way from now on. An advocate who cites such material without checking it first has committed professional misconduct, the Bench held, and a judge or tribunal member who relies on it has committed an equally serious lapse.
The line that will get quoted for years is a blunt one: such a decision has no legal existence at all, the Court said, regardless of how much or how little the fabricated material actually shaped the outcome. Even a fragment of fake material entering the reasoning is enough to require the order to be set aside, because anything less would compromise the integrity of adjudication itself. The Bench also directed the Bar Council of India to constitute a committee – not to regulate AI at large, but to address the narrower problem of advocates placing fake or hallucinated material before a court as precedent, and to prescribe the disciplinary action that will follow a breach.
Why This Reaches Well Beyond the NCLT
It’s tempting to read this as a case about insolvency lawyers and tribunal members. It isn’t. The reasoning is written broadly enough to cover any professional submission that relies on cited authority – a tax opinion quoting a tribunal ruling, an audit memo referencing an accounting standards interpretation, a GST reply citing an advance ruling, a representation before an assessing or adjudicating authority that leans on precedent to make its case. AI-assisted research tools are now routine in most professional offices, ours included, and this ruling attaches a specific, named risk to using them without checking every citation independently. Not a vague caution – misconduct, if the citing party is an advocate, and grounds to set aside the entire order if the citation makes it into a tribunal’s reasoning.
For Chartered Accountants, the exposure sits in two places. The direct one is obvious enough: opinions, submissions, and representations we sign personally, where a fabricated citation is now a documented basis for a misconduct finding rather than just an embarrassing slip. The less obvious one is where we brief counsel or hand over research notes that counsel then relies on. The verification standard this judgment demands has to sit inside our own process at that point – it can’t be something we assume gets caught further downstream.
A Checklist for AI-Assisted Research
- Treat every case citation an AI tool produces as unverified until checked against a primary source – the official reporter, the tribunal’s own site, or a paid database you trust.
- Don’t rely on a citation appearing in an AI-generated summary as proof it exists. Confirm case name, coram, date, and citation number independently – then open the report and verify the actual passage, because a genuine, correctly-cited judgment can still carry a fabricated paragraph or holding grafted onto it. That is exactly what happened here: of the six authorities, three were wholly invented, two were real cases with non-existent paragraphs attributed to them, and one was a real judgment hiding under the wrong case name.
- Where AI tools draft a first cut of research, note down who verified each citation and when, so the verification step can actually be shown if it’s ever questioned.
- Apply the same scrutiny to citations that arrive from junior staff, external counsel, or a client’s own team – in this case the fake material reportedly came from the adjudicating authority’s own research, not from either side’s advocate.
- Put a firm-level policy in place for AI tool use in research and drafting rather than leaving verification to individual habit.
Mistakes Worth Watching For
- Assuming a citation is genuine because it reads in the right format, with a plausible coram and a plausible-sounding holding.
- Carrying a citation forward from an earlier draft or a colleague’s note without re-checking it at the point of final submission.
- Treating this as a litigation-only risk when the same exposure exists wherever a CA’s own opinion or representation cites judicial or tribunal authority.
The Court didn’t ban AI-assisted research, and nothing here requires anyone to stop using these tools. What it requires is that the one step AI tools make easiest to skip – actually verifying the citation – never gets skipped. A firm’s credibility, and in the wrong circumstances its exposure to a misconduct finding, now rests on that single habit holding every time. Not most of the time.
******
The author is a practising Chartered Accountant advising clients on internal audit, regulatory, and tax matters. Views are personal.
