Yes. The Supreme Court of India has held that a judicial or adjudicatory decision contaminated by fake or hallucinated AI-generated precedent cannot stand. Even an “iota” of such material entering the decision-making process can require the decision to be set aside.
In Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. & Anr., 2026 INSC 668, decided on 2 July 2026, the Supreme Court declared “zero tolerance” for the Bar and the Bench citing or relying upon AI-generated precedents without verification. It also expressly held that an advocate citing such judgments without verification commits misconduct.
The ruling does not prohibit lawyers or courts from using artificial intelligence. Its target is narrower and more important: presenting fake, non-existent or hallucinated AI material as genuine legal authority.
What did the Supreme Court actually decide about fake AI citations?
The Supreme Court held that courts must adopt a zero-tolerance approach to producing, citing or using AI-generated precedents without verification.
The controlling observations appear in paragraph 7 of Pooja Ramesh Singh. The Court held three things of immediate practical importance:
An advocate citing AI-generated judgments without verification commits misconduct.
A judge relying on fake or hallucinated AI-generated material as precedent commits a serious lapse.
A decision into which such material enters is liable to be set aside even if the material had only an indirect connection with the ultimate result.
The Court nevertheless expressly protected the rightful use of AI. Its objection was to fake or hallucinated material being presented or relied upon as if it were genuine judicial precedent.
Because Article 141 of the Constitution provides that the law declared by the Supreme Court is binding on all courts within India, the legal principle declared in Pooja Ramesh Singh is not merely an advisory warning.
What happened in Pooja Ramesh Singh v. Jammu and Kashmir Bank?
The case arose from insolvency proceedings concerning Essel Infraprojects Ltd., which had given a corporate guarantee for credit facilities granted by Jammu and Kashmir Bank to Pan India Utilities Distribution Company Ltd.
After defaults occurred, the bank filed an application under Section 7 of the Insolvency and Bankruptcy Code, 2016 before the NCLT, Mumbai. On 28 August 2024, the NCLT admitted the application, appointed an Interim Resolution Professional and declared a moratorium under Section 14. The NCLAT dismissed the subsequent appeal on 11 September 2025.
The problem surfaced when the matter reached the Supreme Court.
Counsel for the appellant pointed out that authorities reproduced in the NCLAT decision and relied upon by the NCLT contained fake or non-existent material. The Supreme Court independently examined them.
Its findings were striking. Of six cited authorities, the problems included:
an existing judgment paired with a wrong citation and a non-existent paragraph;
genuine citations containing paragraphs that did not exist; and
case names and citations that were completely non-existent.
Importantly, the bank filed an affidavit stating that its counsel had not cited those alleged judgments before the NCLT. According to the record before the Supreme Court, the precedents had resulted from the adjudicating authority's own research. The NCLAT then reproduced the authorities without detecting the problem.
The Supreme Court therefore set aside both the NCLT order dated 28 August 2024 and the NCLAT judgment dated 11 September 2025. It restored the Section 7 application to the NCLT for a fresh decision on its merits.
This distinction matters: the Supreme Court did not decide that the underlying insolvency claim itself was invalid. It held that the adjudicatory process had been compromised and therefore required a fresh lawful determination.
What does “AI hallucination” mean in legal research?
An AI hallucination is output generated by an AI system that appears plausible but is factually false or unsupported.
In legal research, hallucinations can take several forms:
Type of error
Example
Completely fictitious case
AI invents a case name and citation that do not exist
Real case, fake paragraph
The judgment exists but the quoted paragraph does not
Correct citation, false legal proposition
AI attributes a rule to a case that never decided that point
Wrong citation
A genuine citation belongs to another case
Fabricated quotation
Words are presented as a judicial quotation although the court never used them
This is why merely checking whether a case name “looks real” is insufficient. Verification must extend to the judgment itself, the cited paragraph and the proposition for which the authority is being used.
Does one fake citation really invalidate an entire decision?
Under the rule stated in Pooja Ramesh Singh, potentially yes.
The Supreme Court deliberately rejected a test under which courts would first ask whether the fake material materially changed the outcome. It stated that the decision must be set aside even if an “iota” of fake or hallucinated material enters the decision-making process.
This is unusually strict because it protects the integrity of adjudication itself.
For businesses involved in tax, insolvency, customs, regulatory or commercial proceedings, the practical lesson is significant. A favourable order is not necessarily secure merely because there were independent arguments supporting it. If the adjudicating authority actually relies on fabricated precedent as part of its reasoning, the order may be vulnerable.
The precise remedy will still depend on the procedural context and the forum hearing the challenge. A party should therefore not assume that every typographical citation mistake automatically nullifies an order. The Supreme Court's ruling concerns fake, non-existent or hallucinated material entering adjudication as legal authority.
Has the Supreme Court applied the zero-tolerance rule again?
Yes.
On 2 September 2026, the Supreme Court applied Pooja Ramesh Singh in Vijay Ghanshyam Gadiya v. Union of India & Anr., 2026 INSC 947.
An Additional Commissioner of Customs, Surat had imposed a penalty of ₹425,27,99,100 under Section 114 of the Customs Act, 1962, concerning alleged misdeclaration of natural diamonds as lab-grown diamonds. The Gujarat High Court had dismissed the challenge to that order.
Before the Supreme Court, it was argued that several judgments and articles used in the adjudication order were AI-generated.
The Supreme Court individually verified the authorities and found case law that was either non-existent or had fake citations. It also found genuine cases that did not contain the legal propositions attributed to them.
Applying Pooja Ramesh Singh, the Court set aside both the customs order and the High Court order and directed fresh adjudication by another officer of the same rank.
The Court added an important principle:
AI can assist adjudication, but assistance cannot replace adjudication.
It described AI as potentially useful as an assistive tool while warning against putting it in the decision-maker's place.
This later ruling is particularly important for businesses because the zero-tolerance principle was applied not merely to a conventional court judgment but to a quasi-judicial administrative adjudication involving a major financial penalty.
Can an advocate be disciplined for submitting fake AI citations?
Yes, but the legal position requires an important distinction.
In Pooja Ramesh Singh, the Supreme Court expressly declared that it is misconduct on the part of an advocate to cite AI-generated judgments without verification.
However, that statement should not be inaccurately translated into “a lawyer is automatically suspended” or “will automatically be disbarred”.
Professional disciplinary consequences are governed by the Advocates Act, 1961.
Under Section 35, where a State Bar Council has reason to believe that an advocate on its roll has been guilty of professional or other misconduct, the matter is referred to its disciplinary committee. After the prescribed hearing process, Section 35(3) permits the disciplinary committee to:
dismiss the complaint;
reprimand the advocate;
suspend the advocate from practice for an appropriate period; or
remove the advocate's name from the State roll.
Accordingly, Pooja Ramesh Singh establishes that unverified fake AI precedent can constitute misconduct. The actual disciplinary punishment, if proceedings are initiated, must still be determined through the statutory disciplinary process.
What did the Supreme Court tell the Bar Council of India to do?
The Supreme Court directed the Bar Council of India (BCI) to constitute a committee to examine advocates submitting fake and hallucinated material as if it were genuine precedent.
The Court asked the BCI to deliberate on preventive guiding principles and the disciplinary consequences that should follow violations.
This direction is separate from the existing disciplinary machinery under the Advocates Act.
As of 1 October 2026, the official material reviewed for this article does not establish that the Supreme Court's separate draft Regulations for Use of Artificial Intelligence in Courts, 2026 have entered into force nationally. They should therefore still be described as draft regulations, not as binding final regulations.
Are the Supreme Court’s 2026 AI Regulations already law?
Not on the material publicly verified for this article.
The Supreme Court's AI Committee published the Draft Regulations for Use of Artificial Intelligence in Courts, 2026 on 3 June 2026. The Supreme Court later extended the deadline for stakeholder comments to 15 July 2026.
The draft itself contemplates commencement through subsequent notification.
That distinction is important for legal accuracy.
The zero-tolerance rule in Pooja Ramesh Singh is part of a Supreme Court judgment and therefore stands independently of whether those draft regulations have been formally brought into force.
The draft framework is nevertheless useful in understanding the judiciary's proposed approach. Among other things, it contemplates human primacy, supervised AI use and permissible applications such as legal research, precedent retrieval, citation verification and document summarisation.
Lawyers and businesses should therefore avoid saying that the draft regulations already impose final nationwide disclosure obligations unless and until the applicable provisions are formally notified.
Did Indian courts warn about ChatGPT before the Supreme Court’s 2026 judgment?
Yes.
One notable earlier decision was the Delhi High Court's Christian Louboutin SAS & Anr. v. M/s The Shoe Boutique–Shutiq, 2023:DHC:6090, decided on 22 August 2023.
The Court warned that AI chatbot responses could contain incorrect responses, fictional case law and imaginative data. It held that ChatGPT could not form the basis for deciding legal or factual questions and said such technology could, at that stage, be used at most for preliminary understanding or research.
The development from 2023 to 2026 is therefore clear:
AI assistance was first treated cautiously; the Supreme Court has now specifically declared a binding zero-tolerance rule against unverified fabricated AI precedents entering adjudication.
Can lawyers still use ChatGPT or other AI tools for legal research?
Yes. The Supreme Court has not banned AI-assisted legal research.
The safer legal distinction is:
Permissible assistance: using AI to identify issues, create research starting points, organise arguments, summarise material or suggest possible authorities for independent verification.
Unacceptable reliance: copying an AI-generated case, quotation, paragraph, statutory provision or proposition into a filing without independently checking the authoritative source.
The Supreme Court itself said its judgment does not affect the rightful use of AI. It requires meaningful human control and verification.
How should an advocate verify an AI-generated citation?
For every authority suggested by an AI system, a lawyer should verify at least four things before filing:
Existence: Does the judgment actually exist?
Citation: Does the citation correspond to that exact case?
Text: Does the paragraph or quotation appear in the authentic judgment?
Ratio: Does the judgment actually support the proposition for which it is being cited?
Where possible, start with the official Supreme Court or High Court judgment, official tribunal/regulator source, or an established legal database containing the actual decision.
Never treat an AI-generated summary of a case as a substitute for reading the relevant portion of the judgment.
What should a business do if an order against it contains suspicious case citations?
A business should have the authorities in the order checked against their original judgments immediately.
A practical review should determine:
whether each case exists;
whether its citation is accurate;
whether the paragraphs quoted actually appear in it;
whether the case states the legal principle attributed to it;
whether the adjudicating authority relied upon the suspect authority; and
what statutory appeal, review, rectification or writ remedy is available.
Timing is critical. Pooja Ramesh Singh does not create a universal new limitation period for challenging an order containing hallucinated citations. The applicable limitation period continues to depend on the particular statute, forum and remedy involved.
Businesses should therefore not delay a challenge while investigating how the hallucination occurred.
Can an AI hallucination help a party reopen a decided case automatically?
No.
The Supreme Court's zero-tolerance rule is powerful, but it does not mean every historical judgment containing a citation error can automatically be reopened indefinitely.
Questions of finality, limitation, maintainability, review jurisdiction, appellate procedure and the nature of the error remain relevant.
A party alleging an AI hallucination should be able to demonstrate precisely:
what material is fake, how it was used in the decision-making process, and what procedural remedy permits the impugned decision to be challenged.
Fact-specific legal advice may therefore be necessary, particularly where the normal appeal or review period has expired.
Practical checklist for businesses and in-house legal teams
Where AI is being used in litigation or regulatory work, adopt a simple rule: AI may generate a research lead; only a verified source may become a legal citation.
Before any pleading, written submission, opinion or regulatory response is finalised:
require the lawyer responsible for the document to verify every authority;
retain copies or links to the judgments actually checked;
verify quotations word-for-word;
check pinpoint paragraphs;
check whether amendments have changed statutory provisions;
never assume that an AI-provided SCC, AIR, neutral or online citation is genuine;
review authorities appearing in orders passed against the company; and
escalate any apparently non-existent precedent immediately.
For high-value matters, a citation-verification checklist should form part of the document-review process.
Frequently Asked Questions
1. What was the Supreme Court's main AI fake-citation case?
The principal ruling is Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. & Anr., 2026 INSC 668, decided on 2 July 2026 by Justices Pamidighantam Sri Narasimha and Alok Aradhe.
2. Is citing a fake AI-generated judgment professional misconduct?
The Supreme Court has expressly held that an advocate citing AI-generated judgments without verification commits misconduct. Any disciplinary penalty must nevertheless follow the process under the Advocates Act, 1961.
3. Is using ChatGPT for legal research illegal in India?
No. The Supreme Court has not prohibited legitimate AI-assisted research. The problem arises when fake or hallucinated material is presented or relied upon as genuine legal precedent without verification.
4. Can a judgment be set aside because of one fake AI citation?
The Supreme Court held that a decision is liable to be set aside even if an “iota” of fake or hallucinated material enters the decision-making process. Whether a particular error falls within that principle must be assessed from the decision and record.
5. Does the rule apply only to lawyers?
No. The Supreme Court described reliance on such material by judges as a serious lapse and declared zero tolerance for both the Bar and the Bench.
6. Does the rule extend to government adjudicating authorities?
The September 2026 decision in Vijay Ghanshyam Gadiya v. Union of India demonstrates its application to a customs adjudication order containing non-existent cases, fake citations and hallucinated legal propositions.
7. Will a lawyer automatically be suspended for a fake AI citation?
No. The Supreme Court has characterised the conduct as misconduct, but suspension is not automatic. Section 35 of the Advocates Act requires disciplinary proceedings, after which the competent disciplinary committee determines the appropriate order.
8. Are the Supreme Court's Draft AI Regulations, 2026 currently binding?
They should still be described as draft regulations on the official material verified as of 1 October 2026. The Supreme Court published them for consultation, and the comment period was extended to 15 July 2026.
9. How can a business detect an AI-hallucinated judgment?
Search for the judgment in the official court source or reliable legal database, confirm the cause title and citation, open the actual judgment, verify the quoted paragraph and then confirm that the case genuinely supports the proposition attributed to it.
10. What should a business do if a tax, insolvency or regulatory order relies on a fake judgment?
Obtain legal review immediately and identify the appropriate appeal, review, rectification or writ remedy. Do not assume Pooja Ramesh Singh creates a separate limitation period; the deadline depends on the governing statute and procedure.