Lawgix AI

Home › Blog

Regulation

What Rules Actually Govern an Indian Advocate's Use of AI (September 2026)

By Akshit Agrawal, Founder, Lawgix AI · Published 26 September 2026 · Figures as at 26 September 2026

None yet. As of 26 September 2026 the Bar Council of India has published no rules on how advocates may use AI for legal research or drafting — 86 days after the Supreme Court expressly directed it to make them, with no deadline attached. What exists instead is three things that are easy to confuse and are not the same: a Supreme Court direction that has not yet produced rules, a BCI circular from 17 July 2026 that touches AI only in the narrow context of deepfakes, and a draft Supreme Court regulation whose public consultation closed on 15 July 2026 and whose outcome has not been published.

That gap matters, because the duty on advocates is already enforceable without any BCI rule at all. The Supreme Court has twice said so this year.

1. What the Supreme Court actually ordered — and what it did not

In Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. & Anr., 2026 INSC 668, decided 2 July 2026, a bench of Justice P. S. Narasimha and Justice Alok Aradhe set aside orders of the NCLT and the NCLAT that rested on non-existent precedents. At paragraph 9 the Court directed:

"we direct the Bar Council of India, being the apex statutory body, to constitute a committee and deliberate on this issue of members of the bar submitting such fake and hallucinated material before the Court as if they are precedents of law. The Bar Council must take up this issue with utmost seriousness, deliberate earnestly, and prescribe a guiding principle to prevent such occurrences, along with the disciplinary action that will follow a violation of the norms."

Two things about that direction are routinely got wrong in the coverage.

First, there is no deadline in it. None. Any article that tells you the BCI has a date to meet has invented one.

Second, it did not create the advocate's duty — it assumed it. The same judgment states, flatly:

"It is a misconduct on the part of an advocate to cite such judgments without verification."

And:

"Such a decision is no decision in the eyes of the law, irrespective of whether such material had a direct or indirect bearing on the decision-making."

So an advocate waiting for the BCI to publish rules before treating verification as a professional obligation has misread the position. The obligation is already stated by the Supreme Court, in terms, and is already characterised as misconduct. The BCI committee is about what disciplinary consequence follows — not about whether the duty exists.

As at the date of writing I have found nothing published by the BCI under that direction. A search is a floor, not proof of absence; if the committee has been constituted and has reported, I have not seen it.

2. What the BCI has actually issued — and why it is not this

The Bar Council of India issued a Circular on Social Media Use dated 17 July 2026, fifteen days after the Supreme Court's direction. It does address artificial intelligence. It does not address the thing advocates actually do with AI.

Reporting the circular, SCC Online records that:

"The BCI has prohibited the creation, publication or circulation of AI-generated images, deepfake videos, voice-cloned audio, face-swapped visuals and other synthetic content falsely depicting judges, courts, advocates, litigants, clients or court proceedings."

That is a real and sensible prohibition. It is also a prohibition on synthetic media about the justice system — not on unverified machine-generated case law in a pleading. An advocate who never makes a deepfake and who files a research memo containing a hallucinated citation has complied with the 17 July circular exactly and is nonetheless squarely within the conduct the Supreme Court called misconduct.

One honest caveat, because it cuts against the tidiness of that reading. Secondary reports of the 17 July package differ: SCC Online reads the AI provisions narrowly, as above; LawBeat has been read as describing a broader disclosure duty extending to AI-generated legal summaries. My working explanation is that the covering directive and an enclosed circular with annexures are not the same instrument. I have not obtained the enclosed circular and its annexures, so I am not asserting either reading. If you have the full package, I would like a pointer.

3. The instrument almost nobody is writing about — and the declaration inside it

On 3 June 2026 the Supreme Court of India, through its AI Committee, released for public consultation a draft titled "Regulations for Use of Artificial Intelligence in Courts, 2026". A first notice invited comments by 20 June 2026; a second notice dated 16 June 2026 extended that deadline to 15 July 2026. (Both dates circulate in the coverage and both are correct — they are two different notices.)

The consultation closed 73 days ago. Nothing published since establishes whether the draft has been finalised, revised or shelved. Until that is public, it is a draft and should be described as one — including here.

It nevertheless matters to practitioners now, for three reasons.

Regulation 43(3) would place a filing obligation directly on advocates:

"Where an AI Tool is used by any party or his legal representative in the preparation or submission of any document, pleading, or evidence, the AI-assisted character of such material shall be disclosed to the Court at the time of submission by way of a duly executed declaration or certificate in the format prescribed under Annexure I"

A prescribed declaration, in a prescribed annexure, attached at the time of submission. If it survives into a final text, it is the single most operationally significant AI rule an Indian litigator will have to comply with — and unlike the BCI direction, it comes with a form.

It is also contested, which is part of why its status is unresolved. Speaking at the International ADR Conference 2026, reported on 13 August 2026, Justice Prathiba M. Singh of the Delhi High Court said of the requirement:

"I think disclosure norms is just going to make lawyers' life more difficult."

and

"They have to OCR documents, e-file documents and at the end of the day, even if you ask for disclosures, they all become templated affidavits, right?"

That is a sitting High Court judge publicly questioning a Supreme Court committee's draft rule. Read it as a signal that Regulation 43(3) in its current form is not a settled outcome.

Regulation 8(1) is the sleeper. It says:

"Accountability for all decisions made by any officer with the assistance of AI shall rest exclusively upon such officer and it shall not be permissible to invoke the outputs of an AI System, the opaqueness of a Black Box system, or the occurrence of hallucination, as a ground for avoiding accountability"

Note the date. 3 June 2026 — three months before Gadiya, and a month before Pooja Ramesh Singh. The "the AI did it" defence was closed off in draft by the Supreme Court's own committee before either judgment had to close it off in fact.

Two further provisions are worth an advocate's attention. Regulation 20(1) would absolutely prohibit reaching a judicial outcome "through Algorithmic Decision-Making alone or solely on the basis of AI-generated information", prohibit AI risk scoring "for any purpose", and prohibit any "undisclosed, opaque, or unexplainable AI System" in a process materially affecting lawful rights or personal liberty. And Regulation 19(1)(d) expressly lists "legal research, precedent retrieval, citation verification and document summarisation" among permitted uses. The permitted-use list is not a vendor endorsement and should not be read as one — but it tells you which direction the Committee thinks the safe uses lie in, and citation verification is on the list.

Sourcing note: the draft and both notices were read from the Supreme Court's own published copies on the Government of India content delivery service and from the Allahabad High Court's posting of the consolidated draft with notice (draft as at 16 June 2026).

4. The part nobody is saying: in both Supreme Court cases, it was the bench, not the bar

Almost every piece written on this in India leads with lawyers beware. Here is what the two judgments actually involved.

In Pooja Ramesh Singh, the non-existent precedents were surfaced by the NCLT — in the Court's own words, "through its own research". They did not come from counsel.

In Vijay Ghanshyam Gadiya v. Union of India & Anr., 2026 INSC 947, decided 2 September 2026, a different bench — Justice Dipankar Datta and Justice Sheel Nagu — set aside a customs penalty of ₹425,27,99,100 imposed by an Additional Commissioner of Customs, Surat, and upheld by the Gujarat High Court, because the adjudicating officer's own order relied on case law that was non-existent or fake-cited. The Court's line:

"AI may well serve as training wheels but entrusting it with the pilot's seat would be both imprudent and dangerous."

And:

"It is necessary for courts to adopt a zero-tolerance mode for producing, citing or using artificial intelligence-generated precedents without verification."

And, critically for the point being made here:

"It is a misconduct on the part of an advocate to cite such judgments without verification. Equally, it is a serious lapse if a judge relies on such fake or hallucinated artificial intelligence-generated material."

The direction at the end of Gadiya is not aimed at a lawyer at all:

"We leave it to the wisdom of the appointing authority to initiate such action against the author of the order dated 8th October, 2025, as may be warranted in the circumstances, in accordance with law."

The author of that order was a revenue officer.

Both things are true at once, and practitioners should hold both. The advocate's verification duty is now expressly stated at the highest level and is characterised as misconduct — take it seriously. And the two cases that state it were both about adjudicators. A ₹425 crore penalty was written on fabricated precedent by the person deciding the matter.

The practical consequence is not reassurance. It is that your exposure is not confined to the tools you buy. Unverified machine output can arrive in the order you are served with — and in Pooja Ramesh Singh the Supreme Court held that such a decision "is no decision in the eyes of the law", whether or not the fake material actually influenced the outcome. That is a ground of challenge. Being able to check a citation quickly is useful whoever produced it.

5. What an Indian advocate should actually do, today

No BCI rule is required for any of this; the case law already supports all of it.

  1. Open every citation before it goes in. Not "check that a link exists" — open the judgment text on a source you would be willing to name to the court. A fabricated citation with a confident-looking link is worse than one without.
  2. Do not treat AI output as a source. Under draft Regulation 8(3), AI output used in court is "advisory in nature" and reasonable care must be taken to verify accuracy before use. That is the draft's standard; it is also plain professional common sense and is what both judgments enforce.
  3. Keep a verification record. If Regulation 43(3) is finalised, you will need to certify AI-assisted character at the time of submission. A practice that already logs what was machine-assisted and what was checked will be able to complete an Annexure I declaration. A practice that does not will be reconstructing it under deadline.
  4. Read the orders you receive with the same suspicion. Both Supreme Court cases arose from the adjudicator's citations, not counsel's.
  5. Do not wait for the BCI. There is no deadline on the direction, it is 86 days old, and the duty is already enforceable.

The cost of getting it wrong already has a number attached. In Deepak Shivkumar Bahry v. Heart & Soul Entertainment, 2026:BHC-AS:828, Justice M. M. Sathaye of the Bombay High Court imposed ₹50,000 in costs on 15 January 2026 where written submissions cited a case that did not exist, observing:

"If an AI tool is used in aid of research, it is welcome; however, there is great responsibility upon the party, even an advocate using such tools, to cross verify the references and make sure that the material generated by the machine/computer is really relevant, genuine and in existence."

Neither Supreme Court judgment imposes costs on a lawyer. The Bombay order remains the case where the price is a figure.

6. What to watch

Summary, as at 26 September 2026: no BCI rules on AI in research or drafting; an unactioned Supreme Court direction 86 days old with no deadline; a BCI circular that covers AI only as synthetic media; and a draft Supreme Court regulation, consultation closed 73 days ago, outcome unpublished. The advocate's verification duty does not depend on any of them and is already enforceable as misconduct.

Sources

This article is general information about the regulatory position as at 26 September 2026, not legal advice.

Lawgix AI builds legal AI for Indian law firms and in-house teams — research, document review and drafting grounded in Indian Acts and judgments, with citations you can check.

See how it works

← All articles