Yes — an Indian court can reject a petition that relies on AI/ChatGPT-generated citations, and citing fake or unverified AI-generated case law can amount to professional misconduct under Section 35 of the Advocates Act, 1961. Courts have dismissed such filings, and in 2026 the Supreme Court held that relying on non-existent AI-generated judgments "would be a misconduct and legal consequence shall follow," directing the Bar Council of India to frame norms. What is not banned is using AI to draft at all — the misconduct lies in filing the output without independently verifying every citation and proposition against a primary source.
That distinction is the whole subject of this page. A junior advocate who uses ChatGPT to produce a first draft and then checks every authority on Indian Kanoon or SCC has done nothing wrong. A junior who files the draft as-is, trusting cases the model invented, has exposed a client to dismissal and themselves to a disciplinary complaint. Below is exactly what the courts have said, what can happen to you, and the single discipline that separates the two outcomes.
Professional misconduct, in this context, means presenting to a court — without verification — case law, quotations or propositions that an AI tool fabricated, thereby misleading the court and breaching the advocate's duty of candour.
Is using AI or ChatGPT to draft legal documents itself banned in India?
No. There is no statute, Bar Council of India rule, or Supreme Court order that prohibits an advocate from using ChatGPT, Claude, Gemini or any research tool to draft pleadings, notices or research notes. The courts' concern has never been the tool; it has been unverified output filed as if it were verified.
The Supreme Court itself drew this line. In Gummadi Usha Rani & Anr. v. Sure Mallikarjuna Rao & Anr. (SLP (C) No. 7575 of 2026, order dated 27 February 2026, Narasimha and Aradhe JJ.), the Court was careful to target the reliance on fabricated material, not AI use in general. The practical rule that emerges is simple and evergreen:
An advocate may use AI to assist drafting and research, but remains personally and fully responsible for the accuracy of every citation and statement in a filing — exactly as they are for work produced by a junior or a clerk.
This is why this page frames the question around the permanent professional-responsibility framework rather than any single news story: the duty to verify predates AI and will outlast any particular tool. If you want the step-by-step method, see our guide on how to verify AI-generated case citations before filing in India.
What exactly did the Supreme Court say in 2026?
Two 2026 orders by the same Bench (Justices P.S. Narasimha and Alok Aradhe) set the current position.
1. Gummadi Usha Rani (27 February 2026) — the judicial-side warning. The Court took cognizance of a trial court order dated 19 August 2025 that had relied on four decisions the petitioners said were non-existent and fake — one citation even bore the impossible year "1071," a tell-tale sign of AI hallucination. The Bench held that a decision built on "AI generated non-existing, fake or synthetic alleged judgments" is not a mere error in decision-making: in its words, "it would be a misconduct and legal consequence shall follow." The Court issued notice to the Attorney General, the Solicitor General and the Bar Council of India to examine the wider implications.
2. Pooja Ramesh Singh v. Jammu & Kashmir Bank (2026 INSC 668, 2 July 2026) — the advocate-side rule. The same Bench set aside insolvency orders of the NCLT and NCLAT that had rested on precedents which did not exist. The Court observed that it is misconduct on the part of an advocate to cite such judgments without verification, and again flagged the need for the Bar Council of India to frame disciplinary norms for AI-assisted practice.
High Courts have echoed the same theme. In a widely reported matter, the Delhi High Court (Kathpalia J.) confronted a plea stuffed with fake ChatGPT-generated quotes and cases; the plea was permitted to be withdrawn after the Court observed that such use of AI was "impermissible" and a matter of serious concern. The Bombay High Court has similarly cautioned lawyers against filing AI-generated authorities without checking them.
The takeaway is consistent across all of them: the sanction attaches to the fabrication reaching the court unverified, not to the use of AI.
What are the consequences under Section 35 of the Advocates Act, 1961?
Section 35 is the disciplinary engine. When a State Bar Council "has reason to believe that any advocate on its roll has been guilty of professional or other misconduct," it must refer the case to its disciplinary committee (s.35(1)). After hearing the advocate and the Advocate-General, that committee may, under s.35(3):
| Order under s.35(3) | Effect |
|---|---|
| (a) Dismiss the complaint | No action; proceedings filed |
| (b) Reprimand the advocate | Formal censure on record |
| (c) Suspend from practice | For a period the committee deems fit; during suspension the advocate cannot practise in any court or before any authority in India (s.35(4)) |
| (d) Remove the name from the State roll | Effectively ends the right to practise |
An order of the State disciplinary committee is appealable to the Bar Council of India under Section 37, and onward to the Supreme Court under Section 38. Note that the Act does not itself define "misconduct" — courts have read it broadly to cover any conduct that falls short of an advocate's professional obligations, which is why filing fabricated authorities fits squarely within it.
Section 35 is not the only exposure. In the same breath a court can:
- Dismiss or disregard the petition (or the offending portions), costing the client the matter;
- Impose costs on the party or counsel for wasting judicial time;
- Initiate contempt proceedings where the fabrication amounts to interference with the administration of justice; and
- Record adverse remarks that follow the advocate into future appearances.
In short: the realistic worst case is not just a lost case — it is costs, a disciplinary reference under s.35, and lasting reputational damage, any of which is avoidable by verification.
Will the court reject the whole petition, or can I fix it?
It depends on when the problem surfaces and how you respond. In practice:
- If you catch it before filing: no consequence at all — you simply correct the draft. This is the only safe category, and it is entirely within your control.
- If opposing counsel or the judge catches a fake citation: expect, at minimum, the citation to be struck and your credibility for the rest of the hearing to suffer. Courts have allowed petitions to be withdrawn rather than decided on merits, which saves the client's substantive rights but is an embarrassing and costly detour.
- If you are confronted in court: candour is the only defence. Acknowledge the error immediately, withdraw the citation, and offer to verify and re-submit. Doubling down — insisting a fabricated case is real — is what converts a mistake into misconduct.
There is no "the AI did it" defence. Responsibility for a filing rests on the advocate who signs and files it, full stop.
How do I use AI safely? The one discipline that keeps you safe
Every court warning reduces to a single rule: verify before you file. Operationally, that means:
- Treat every AI-provided citation as unverified until you open the judgment. Pull it on Indian Kanoon, the Supreme Court's own portal, the relevant High Court site or SCC/Manupatra. If you cannot find it, it does not exist — delete it.
- Read the paragraph, not just the headnote. AI frequently cites a real case for a proposition it never decided. Confirm the ratio actually supports your point.
- Check the citation format and year. Impossible years (like the "1071" in Gummadi Usha Rani), mismatched reporters, or a case number that doesn't resolve are classic hallucination markers.
- Verify statutory provisions on India Code. Post-2023 criminal law moved from the IPC/CrPC to the BNS/BNSS/BSA; AI trained on older text routinely cites repealed sections.
- Keep a verification trail. A one-line note against each authority ("verified on Indian Kanoon, [date]") protects you if a filing is ever questioned.
For finding authorities that genuinely match your facts — rather than ones a model guessed at — see how to find case law matching your facts in India. Hindi-speaking readers will find the same analysis of court rejection and misconduct in क्या AI/ChatGPT से बनी याचिका कोर्ट खारिज कर सकता है, and a practical walkthrough of AI-assisted research in AI से कानूनी शोध कैसे करें. And if you are still deciding what to use, see our comparison of AI legal drafting and research tools in India — free vs paid, and what each actually does.
Does this apply to law students, interns and litigants-in-person?
The disciplinary machinery of Section 35 applies only to enrolled advocates — a student or an in-person litigant cannot be "struck off." But the court's power to reject a filing, impose costs, and act for contempt applies to everyone. A litigant-in-person who files AI-fabricated citations will have them disregarded and may be penalised in costs. For students, the risk is academic and reputational — moot court and internship submissions built on fake authorities fail hard and are easily caught. The verification discipline above is identical for all three groups.
How Urava helps
Urava is built around exactly the discipline the courts now demand. It produces court-ready, citation-backed legal research memoranda for Indian law in about ten minutes — from a typed question or a scanned vernacular document — with every authority tied to a real, checkable source rather than a model's guess. That means you get the speed of AI drafting without inheriting the risk of unverified, hallucinated case law, and you keep a clean trail of what was relied on. You can ask in English, Hindi or Malayalam over WhatsApp, and the free tier gives you three researches to test it against a matter you already know the answer to. Start free at urava.app/register and see the difference verification makes.
Frequently Asked Questions
Is it illegal for a lawyer to use ChatGPT in India?
No. No Indian law or Bar Council rule bans advocates from using ChatGPT or other AI tools to draft or research. What is sanctionable is filing AI output — especially citations — without independently verifying it. The tool is permitted; unverified reliance is the risk.
Can a court reject my petition just because AI helped write it?
Not for the mere fact of AI assistance. A court rejects a petition when it contains fabricated, non-existent or misquoted authorities, or when the drafting is misleading. A verified, accurate AI-assisted petition stands on exactly the same footing as one drafted by hand.
What did the Supreme Court decide about AI-generated fake citations?
In Gummadi Usha Rani (27 February 2026) the Court said reliance on fake AI-generated judgments "would be a misconduct and legal consequence shall follow," and in Pooja Ramesh Singh v. J&K Bank (2026 INSC 668, 2 July 2026) it held that citing unverified AI-generated precedents is misconduct by an advocate. It asked the Bar Council of India to frame norms.
What punishment can an advocate face for filing fake AI citations?
Under Section 35(3) of the Advocates Act, 1961, a State Bar Council disciplinary committee may reprimand the advocate, suspend them from practice for a period, or remove their name from the roll. Separately, a court can impose costs, dismiss the filing, or initiate contempt.
How do I make sure an AI-generated case is real before filing?
Open the full judgment on Indian Kanoon, the Supreme Court/High Court portal, or SCC. Confirm the case exists, that the paragraph actually supports your proposition, and that the citation year and number resolve. If you cannot locate it, treat it as fabricated and remove it.
Is "the AI made the mistake" a valid defence?
No. Responsibility for a filing rests entirely with the advocate who signs and files it, the same as for work done by a clerk or junior. Courts have made clear that AI assistance does not dilute the advocate's personal duty to verify.