- India's Supreme Court warns against AI-generated fake legal precedents in courts
- The court urges Bar Council to create AI use policy for the legal profession
- Judges and lawyers must verify AI outputs and not treat them as authoritative
India's Supreme Court has delivered one of its clearest warnings on the risks of generative artificial intelligence in the legal system, saying there will be “zero tolerance” for hallucinated precedents and fabricated authorities being presented before courts. The Supreme Court has asked the Bar Council of India to frame a policy on AI use in the legal profession.
The judgment does not reject AI, but places responsibility on lawyers and judges to verify what AI tools produce. The bench repeatedly stressed the need to distinguish “between fact and fiction, between fact and analysis, what is real and what is unreal” and cautioned against treating AI-generated outputs as legal authorities without independent verification.
The court compared the introduction of fake AI-generated precedents into the justice system to the release of methyl isocyanate, the toxic gas associated with the 1984 Bhopal disaster. The comparison underscored the judges' concern that fabricated authorities could spread through judgments, legal databases and future litigation if left unchecked.
The warning comes even as India's judiciary expands its use of AI. The Supreme Court already uses tools such as SUPACE for legal research assistance, SUVAS for translation of judgments and AI-powered transcription systems. Last month, the court's AI Committee released draft Regulations for Use of Artificial Intelligence in Courts, 2026, proposing a broader framework for AI adoption while making clear that judicial decision-making must remain in human hands.
The immediate dispute before the court involved hallucinated legal authorities. But several lawyers said the larger concern was that the fabricated citations survived multiple layers of scrutiny before being detected.
The fake authorities were not caught during legal research, courtroom arguments, proceedings before the National Company Law Tribunal or appellate review before eventually reaching the Supreme Court.
“The key distinction here is between fact and fiction, between fact and analysis, what is real and what is unreal,” said Arun Prabhu, Partner and Co-Head, Digital+, TMT, at Cyril Amarchand Mangaldas.
“Where a lawyer (or in this case, a tribunal) cites a case (or a proposition in a case) that does not exist as precedent, this is a misstatement of fact, not an error of analysis.”
Prabhu said the judgment highlights that the professional duty of lawyers does not change simply because AI is involved. A lawyer who cites a non-existent case would face consequences whether the mistake arose from carelessness, intention or reliance on an AI tool, he said.
“The judgement calls out the consequences of not exercising professional diligence while relying on AI citation, as well as the legal consequences on judgement that rely on hallucinated precedent. It does not change the underlying duty of care, but illustrates the consequences of its breach in specific circumstances,” Prabhu added.
The ruling follows other recent Indian cases in which lawyers cited authorities that did not exist, including KMG Wires v. NFAC Delhi and Greenopolis Welfare Association v. Narender Singh. Those matters reflected a problem courts around the world are beginning to confront: AI systems can generate convincing-looking citations that are entirely fictional.
For Shiv Sapra, Partner at Kochhar & Co, the judgment exposes a weakness in the legal system, which allowed hallucinated precedents to slip through multiple layers of judicial scrutiny.
“That highlights an overreliance on secondary research and an assumption that every cited authority has already been verified. AI did not create the underlying problem; it amplified an existing weakness in legal research practices by making fabricated material appear highly convincing,” Sapra explains.
The Supreme Court's approach is consistent with broader policy discussions within the Indian judiciary. Last year, the Kerala High Court introduced a policy requiring judges to independently verify AI-generated material, maintain audit trails and remain personally responsible for the contents of their orders.
The Supreme Court's November 2025 white paper on AI and the judiciary also supported AI-assisted research and court administration while emphasising verification, human oversight and accountability.
The draft AI regulations released in June take the same approach further. They prohibit AI from deciding cases, determining bail eligibility, assessing witness credibility, assigning risk scores or performing other adjudicatory functions, while encouraging AI-assisted research, translation, accessibility services, transcription and case management.
One unresolved question is how lawyers themselves should be regulated when using AI tools. The Advocates Act and Bar Council of India rules already impose duties of competence, candour and diligence, but they were drafted before generative AI became widely available.
“This may well develop into an Indian version of a duty of AI competence,” said Ankit Sahni, Partner at Ajay Sahni & Associates. “The professional duty is therefore not to avoid AI altogether, but to use it with human supervision, source verification and full accountability.”
“AI can be used responsibly only when lawyers understand both its utility and its risks, especially hallucinated cases, fabricated quotations and overconfident but incorrect legal propositions,” Sahni says.
Shaurya M. Tomar, Senior Partner at Chugh Universal Legal, said the ruling could shape future professional standards.“The judgment reinforces a fundamental principle of legal practice: AI is an effective aid for legal research, but it cannot replace professional diligence,” Tomar said.
India's courts are preparing for wider use of AI, and the judiciary's draft regulations make clear that technology is expected to play a growing role in research, administration and accessibility.
As Sapra put it: “Competence today is no longer confined to knowing the law; it also includes knowing when technology can be trusted and, more importantly, when it cannot.”
ALSO READ: Gujarat HC Upholds Death Penalty For 38, Life Term For 11 In 2008 Ahmedabad Blasts
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
