Last updated: 12 November 2026
Using ChatGPT or a similar tool to start legal research is tempting: it is fast, fluent and it sounds sure of itself. In India, that confidence has already cost lawyers and even a trial court dearly. This guide sets out the mistakes that matter most when generative tools are used for legal work, and the correct position under Indian law and court practice as of the date above.
The short version: these tools can help with structure, plain language summaries and first drafts, but they are not a source of law. Every authority must be found and read in its original form by a human who then signs for it.
Learn this properly. Lexfolio offers AI for Lawyers and Law Students, a self-paced certificate course on using AI tools in legal research and drafting, for lawyers and law students. Course price: Rs 499.
Mistake 1: Citing a case the tool “found”
Language tools predict plausible text. They do not look up judgments, so they can produce a case name, a court, a year and a neat holding for something that was never decided. This is not a theoretical risk in India. On 17 February 2026 a Supreme Court bench of the Chief Justice of India, Surya Kant, Justice Joymalya Bagchi and Justice B.V. Nagarathna said that it had begun noticing pleadings that seemed to have been prepared with AI platforms without adequate human scrutiny, and Justice Nagarathna referred to a matter citing a precedent called “Mercy vs Mankind” that does not exist. The Chief Justice also said that in proceedings before another judge, the precedents cited had never existed, and that some filings quoted passages from real Supreme Court rulings that do not appear in them.
In Gummadi Usha Rani v. Sure Mallikarjuna Rao (SLP (C) No. 7575 of 2026), a trial court had relied on four decisions the petitioners said were non-existent and AI generated. On 27 February 2026 the Supreme Court said that a decision based on such fake judgments is not an error in decision making, and that relying on them would be misconduct with legal consequences; it issued notice to the Attorney General, the Solicitor General and the Bar Council of India (BCI) and appointed a senior advocate as amicus curiae. At the hearing reported on 5 May 2026 the Court said that parties cannot rely on AI-generated material and then apologise, and asked the BCI to constitute a committee of experts to examine the issues. It also said it was not seeking to prohibit the use of AI.
Mistake 2: Trusting a summary of a real case
A tool may name a genuine case and still misstate it: the wrong bench strength, the wrong outcome, a passage that is not in the judgment, or a proposition the court actually rejected. Read the judgment. Look at who sat, the date, the question decided, the reasoning and the direction. If you intend to quote, copy the words from the judgment itself and give the paragraph number. A summary is a pointer to where to look, never the authority itself.
Mistake 3: Using old or wrong law
A tool trained mostly on older text may answer in terms of the Indian Penal Code, the Code of Criminal Procedure or the Evidence Act. Those were replaced from 1 July 2024 by the Bharatiya Nyaya Sanhita, 2023 (BNS), the Bharatiya Nagarik Suraksha Sanhita, 2023 (BNSS) and the Bharatiya Sakshya Adhiniyam, 2023 (BSA). Section numbers changed: for example, cheating is section 318 BNS, cheating by personation is section 319 BNS, and the privilege for professional communications with an advocate is section 132 BSA. Other laws have changed too, including the Labour Codes in force from 21 November 2025. Always check the current text of the statute on the India Code portal or the relevant Gazette, and check whether your authority was decided under the old or the new provision.
Mistake 4: Pasting client material into a public tool
Advocates owe clients confidentiality, and section 132 of the BSA bars an advocate from disclosing communications made in the course of professional engagement without the client’s express consent, subject to defined exceptions. In Re: Summoning Advocates who give legal opinion or represent parties during investigation (31 October 2025, 2025 INSC 1275) the Supreme Court held that this privilege covers litigation and non-litigious work, and that an investigating officer cannot directly summon an advocate to disclose case details unless an exception under section 132 applies. The same judgment noted that in-house counsel are not “advocates” under the Advocates Act, 1961 and so cannot claim the section 132 privilege, though they may receive some protection under section 134.
None of this approves or forbids a particular tool. But sending a client’s confidential papers to an outside service is a disclosure that the client has not agreed to, and it may place privileged material outside your control. The safe practice is to use approved tools under written terms, remove names and identifiers, or use no tool at all for the most sensitive files.
Mistake 5: Letting the tool do the judgment
The tool can suggest arguments. It cannot weigh the facts of your case, the temperament of the bench or the client’s interests. The Kerala High Court’s policy on the use of AI tools in the district judiciary, issued on 19 July 2025, shows how courts themselves are drawing the line: such tools must not be used for decision making or legal reasoning or to arrive at findings, reliefs or orders, outputs including translations must be carefully checked, and cloud based tools such as ChatGPT, Gemini, Copilot and Deepseek are to be avoided unless approved. That policy applies to Kerala’s district judges and their staff, not to advocates, but it is a useful statement of principle for anyone in the system.
Mistake 6: Thinking there is no rule for lawyers
There is no AI specific code for advocates yet, but there are rules in force. Professional misconduct under the Advocates Act, 1961 and the Bar Council of India Rules can reach an advocate who files fabricated authorities, whatever the source. The Bar Council of India has told Parliament, as reported in June 2026, that advocates who file AI-generated fake citations, fabricated quotations or non-existent judgments cannot escape professional liability by blaming technology. Courts may also impose costs, call for explanations or refer the matter to the Bar Council.
Separately, a draft policy on AI use in courts was issued by the Supreme Court’s AI Committee on 3 June 2026 for public comment, with comments invited until 15 July 2026, and would be notified by High Courts if adopted. Check the current status and any High Court practice direction before you file.
Mistake 7: Not telling the client, or billing as if the tool did not exist
If a client has asked whether AI is used, or the engagement letter says nothing, be open. Explain what the tool is used for, that a lawyer reviews everything, and that confidential data is protected. Fees should reflect work actually done by you. Clients who find out later that unchecked output went to court will be less forgiving than clients who were told at the start.
Mistake 8: Skipping the “is it still good law” check
Even a real, correctly cited judgment may have been overruled, distinguished, or limited by a larger bench, or stayed in review. A language tool will not tell you reliably. Run the authority through a citator in a recognised law report database, read the later cases that cite it and check the date against any statutory change.
A quick test before you rely on any output
- Can I open the judgment or the statute section on an authoritative source?
- Have I read the paragraph, not just the headnote or the tool’s summary?
- Is the quote verbatim, with a paragraph number?
- Was it decided under the old law or the new one?
- Is it still good law?
- Did I keep client confidences?
If any answer is no, do not file it.
Learn this properly. Lexfolio offers AI for Lawyers and Law Students, a self-paced certificate course on using AI tools in legal research and drafting, for lawyers and law students. Course price: Rs 499.
Frequently asked questions
Has the Supreme Court banned the use of AI by lawyers?
A: No. At the hearing reported on 5 May 2026 it said it was not seeking to prohibit the use of AI, but it made clear that parties cannot rely on AI-generated material and then apologise for errors.
Can a lawyer be punished for citing a fake case generated by a tool?
A: Citing a non-existent authority can amount to professional misconduct and can lead to costs and Bar Council proceedings. The Bar Council of India has said responsibility cannot be passed on to technology.
Is a tool’s summary of a judgment safe to use if the case name is real?
A: No. A real case name can be paired with a wrong or invented holding or quotation. Read the judgment and quote from it.
Does privilege protect what I paste into a chatbot?
A: Section 132 of the BSA protects confidential professional communications from compelled disclosure by the advocate, but a voluntary disclosure to a third party service can create its own risk. Use approved tools and anonymise.
This article is general information, not legal, financial or tax advice for any specific case. Lexfolio is a learning resource and does not provide legal services.