Supreme Court Draws a Red Line on AI-Hallucinated Legal Citations

In a significant 2026 ruling, the Supreme Court set aside insolvency orders after discovering that the adjudicatory process had relied on non-existent judgments and fabricated passages. The decision makes one point unmistakable: artificial intelligence may assist legal research, but every authority and citation must be independently verified.

Supreme Court Draws a Red Line on AI-Hallucinated Legal Citations

Generative artificial intelligence is increasingly being used for legal research, drafting, translation and document review. Its speed is attractive, but its ability to produce convincing yet false information creates a serious risk when the output is used in court proceedings.

The Supreme Court of India addressed this issue directly in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., 2026 INSC 668, decided on 2 July 2026.

The Court held that reliance on fake, non-existent or hallucinated precedents contaminates the adjudicatory process. A decision influenced by such material cannot be sustained merely because the false citation may not have determined the final result.

What Happened in the Case?

The dispute arose from insolvency proceedings against Essel Infraprojects Ltd., which had provided a corporate guarantee for credit facilities granted to another company.

The Mumbai Bench of the National Company Law Tribunal admitted an application under Section 7 of the Insolvency and Bankruptcy Code on 28 August 2024. The National Company Law Appellate Tribunal subsequently dismissed the appeal on 11 September 2025.

When the matter reached the Supreme Court, it was pointed out that six authorities referred to in the tribunal proceedings were problematic. Some of the cited cases did not exist. In other instances, the case or citation was genuine, but the paragraph or legal proposition attributed to the judgment could not be found in the actual decision.

The Supreme Court’s verification found that two citations were genuine but had been attributed with non-existent passages, one citation belonged to an entirely different judgment, and the remaining three citations did not exist. The Court also noted that the bank’s counsel stated that these authorities had not been supplied by the bank’s lawyers and appeared to have entered the tribunal’s reasoning through its own research.

The Supreme Court's Decision

The Supreme Court set aside the NCLT order dated 28 August 2024 and the NCLAT order dated 11 September 2025. It restored the Section 7 application to its original number and directed the NCLT to decide the case afresh on its own merits.

Importantly, the Supreme Court did not decide whether the insolvency application should ultimately succeed. Its concern was that the adjudicatory process itself had been compromised by reliance on authorities that were not genuine.

Why Fake Citations Are More Serious Than an Ordinary Mistake

An incorrect interpretation of a real judgment can be challenged through ordinary appellate review. A fabricated judgment presents a different problem: the supposed legal source never existed in the first place.

If a court, tribunal or lawyer treats invented material as binding precedent, the reasoning appears legally supported even though its foundation is false. The opposing party must spend time and resources proving that the cited authority is imaginary or that the quoted paragraph is absent from the real decision.

The Supreme Court therefore held that even a minimal amount of fake or hallucinated material entering the decision-making process can invalidate the adjudication. It declared a zero-tolerance approach for both the Bar and the Bench in relation to unverified AI-generated precedents.

Did the Supreme Court Prohibit the Use of AI?

No.

The judgment expressly clarified that it does not affect the legitimate use of artificial intelligence. The concern is not the mere use of an AI tool; it is the presentation or reliance upon false AI-generated material as though it were an authentic court precedent.

AI can assist lawyers with:

• Developing research keywords.
• Preparing preliminary summaries.
• Translating documents.
• Structuring chronologies.
• Identifying potential issues.
• Creating first drafts.
• Comparing large volumes of text.

However, an AI-generated answer is not a legal authority. A case name, citation, quotation or statutory provision generated by AI must be checked against the original judgment, official court record, statutory text or a reliable legal database.

Professional Responsibility Remains with the Lawyer

A lawyer cannot defend a false filing by stating that the error was made by software.

The Supreme Court directed the Bar Council of India to constitute a committee to examine the submission of fake and hallucinated authorities by members of the Bar. The committee was asked to formulate guiding principles to prevent such occurrences and consider the disciplinary consequences that should follow a violation.

The ruling reinforces a fundamental principle: technological assistance does not dilute professional responsibility. The advocate who signs, settles or files a document remains responsible for its accuracy.

The same principle applies within law firms. A partner or supervising advocate should not assume that a junior lawyer has verified authorities merely because a draft contains formal-looking case citations.

A Safer Protocol for AI-Assisted Legal Research

Law firms, legal departments and individual practitioners should adopt a written verification process.

1. Verify the Existence of the Case

Search the case name and citation on the relevant court website or a recognised legal database. Confirm that the parties, court, date and citation correspond with one another.

2. Open and Read the Full Judgment

A headnote, online summary or AI-generated extract is not a substitute for the complete judgment. The lawyer should examine the facts, issues, reasoning, final order and any separate or dissenting opinions.

3. Verify the Exact Paragraph

Every quotation and paragraph reference should be checked against the original text. AI tools may attach a genuine proposition to the wrong case or invent a paragraph within a real case.

4. Check Subsequent Treatment

Confirm whether the judgment has been overruled, distinguished, stayed, reviewed or limited by a later decision.

5. Maintain a Research Record

The file should contain a copy or link to every material authority relied upon, together with the relevant paragraph. This allows another member of the team to repeat the verification.

6. Apply Human Legal Judgment

Even where the case and quotation are accurate, a lawyer must decide whether the authority applies to the client’s facts and whether it is binding, persuasive or distinguishable.

Implications Beyond Litigation

The danger of AI hallucination is not confined to court pleadings. Businesses are using generative AI to prepare employment policies, contracts, compliance reports, due-diligence notes and regulatory responses.

A false statutory requirement, incorrect deadline or imaginary regulatory circular can expose a business to financial and reputational harm even when no case has been filed.

Organisations using AI for legal or compliance work should therefore ensure that:

• AI outputs are reviewed by a qualified person.
• Confidential client or business information is not entered into an unapproved public system.
• Source documents are preserved.
• Important legal conclusions are verified independently.
• Internal policies clearly allocate responsibility for final approval.

Conclusion

The Supreme Court’s message is not that legal professionals must reject technology. The message is that speed cannot replace verification.

Artificial intelligence can identify possibilities, organise information and accelerate routine work. It cannot certify that a case exists, determine whether a quotation is accurate or accept professional responsibility for a filing.

In legal practice, an answer that sounds authoritative is not enough. It must be traceable to a genuine and properly understood source.

Avyaksham Legal LLP assists clients with technology-law compliance, AI-governance policies, legal-risk assessments and the review of AI-assisted contractual and regulatory documentation.

Disclaimer: This article is for general information only and does not constitute legal advice. Legal authorities and citations must be verified against the official judgment or an authorised legal database before being relied upon.