A misconduct line, drawn deliberately
On May 5, 2026, a Supreme Court bench of Justices P.S. Narasimha and Alok Aradhe directed the Bar Council of India (BCI) to constitute a committee of independent experts to examine the use of artificial intelligence in litigation and judgment-drafting, and to file a report before the court ahead of a hearing listed for May 26. The order extends the bench's earlier declaration of February 27, 2026: that "a decision based on such non-existent and fake alleged judgments is not an error in the decision-making process. It would be a misconduct and legal consequence shall follow."
What the court conspicuously did not do is ban the technology. "We are not suggesting people not to use AI," the bench said on May 5. "But we should have control over the information." That restraint is the most important line in the order, and it should anchor whatever guidelines the BCI panel eventually proposes.
The case that forced the question
The trigger was an Andhra Pradesh trial court order dated August 19, 2025. In dismissing objections to an advocate commissioner's report, the court relied on four Supreme Court "precedents" — among them Subramani v. M. Natarajan and Lakshmi Devi v. K. Prabha — that simply do not exist. They were textbook AI hallucinations: plausible citations, real-sounding parties, fictional law.
It was not a one-off. In December 2024, the Bengaluru bench of the Income Tax Appellate Tribunal issued an order in Buckeye Trust v. PCIT citing Supreme Court and Madras High Court judgments that were never decided; the tribunal recalled the entire order within days once the fabrications were exposed. In September 2025, a litigant before the Delhi High Court invented paragraphs from Raj Narain v. Indira Nehru Gandhi (1975) — a judgment that contains only 27 paragraphs. In January 2026, the Bombay High Court imposed ₹50,000 in costs on a party for filing AI-generated fake citations, the judge noting the "obvious giveaway features of a raw AI output."
The strongest case for the committee
The case for the BCI panel is real, and worth stating plainly. Judicial legitimacy rests on verifiability: a ruling is only as trustworthy as the authorities it stands on. A fabricated citation that escapes detection can be quoted in later orders, compounding the error across the system. With India's courts carrying tens of millions of pending matters and overstretched judges increasingly tempted to use AI for first-draft research, the risk of contamination is structural, not anecdotal. Clear, uniform expectations — rather than ad hoc cost orders that vary by bench — would give lawyers and judges a predictable standard and protect litigants who never consented to having their cases decided on invented law.
Existing law already reaches the harm
Here is the pro-innovation point the committee should keep in front of it: India does not need a new AI statute to punish this conduct, because the law already targets the conduct, not the tool. Section 35 of the Advocates Act, 1961 empowers State Bar Council disciplinary committees to reprimand, suspend, or remove advocates for professional misconduct. Submitting fabricated authority is close to a paradigm case, and serious instances can additionally attract contempt. The misconduct framing the bench adopted on February 27 is not a gap in the law — it is the law working.
The United States reached the same destination without new legislation. In Mata v. Avianca (June 22, 2023), Judge P. Kevin Castel sanctioned two attorneys $5,000 under the existing Rule 11 after they filed a brief full of ChatGPT-invented cases. No "AI law" was required; a verification duty enforced through ordinary professional-responsibility rules did the work. India's Section 35, its contempt powers, and the court's own misconduct declaration already form an equivalent toolkit.
The overcorrection to avoid
The danger now is that the committee reaches for remedies disproportionate to the harm. Three are worth flagging. A blanket ban on AI in chambers would be unenforceable and self-defeating, foreclosing genuinely useful applications — translation across India's languages, transcription, document summarisation — that ease the very backlog the courts are drowning in. A disclosure regime so heavy that every AI-assisted keystroke must be logged would deter cautious, beneficial use while doing nothing to stop the careless. And proposals floated in the wider debate to mandate a single state-built "sovereign" legal model as the only sanctioned research tool would trade a manageable verification problem for a far worse one: a slow-moving government monopoly, locked to one vendor and one update cycle, that practitioners would route around anyway.
The proportionate design is narrow and technology-neutral. Require the human who signs a filing or an order to certify that every citation was checked against a primary source — the official reporter or the court's own record. Keep liability squarely on that human. Treat the AI tool as no different from an unreliable junior: useful, but never a substitute for the signer's own verification. That approach polices the harm, preserves innovation, and needs no new prohibition.
What to watch
The BCI committee's report — and the bench's response at the next hearing — will set the tone for how a major common-law jurisdiction governs AI in adjudication. If it codifies a verification-and-certification duty and resists tool bans, India will have produced a template worth exporting. The court's instinct so far — punish the misconduct, keep the technology — is the right one.