AI-Generated Authorities and the Integrity of Judicial Proceedings: Reconstructing the Lawyer's Duty of Verification in India
TL;DR
A tiered, source-anchored verification standard is proposed, keyed to source authoritativeness, procedural stage, and research-responsibility allocation within a legal team, capable of catching fabricated authorities before a judge relies on them rather than after an entire tier of appeal already has.
Abstract
Between December 2024 and July 2026, Indian tribunals and trial courts relied on judicial precedents that did not exist, generated by a large language model and accepted without verification; in one instance, a High Court that itself recognised such a fabrication on revision nonetheless declined to treat it as fatal to the outcome. The pattern culminated in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., where the Supreme Court set aside a National Company Law Tribunal order and its confirmation in appeal, both resting on fabricated precedents, and declared that citing AI-generated authorities without verification is professional misconduct. This paper argues that the Supreme Court’s response, while doctrinally significant, has so far declared a duty of verification without defining its content: it does not specify what verification requires, against which sources, how it should be documented, or how responsibility should be apportioned between an Advocate-on-Record, briefing counsel, and the junior associates who typically conduct the underlying research. The paper traces the existing architecture of the advocate’s professional obligations under Sections 35 and 36 of the Advocates Act, 1961 and the Bar Council of India’s Standards of Professional Conduct and Etiquette, and shows that this architecture, framed before generative artificial intelligence existed, supplies only generalised duties of diligence and candour rather than a specific, source-anchored protocol. Drawing on the doctrinal lineage from Gummadi Usha Rani v. Sure Mallikarjuna Rao through Pooja Ramesh Singh, on the Supreme Court’s draft Regulations for Use of AI in Courts, 2026, and on comparative practice under Mata v. Avianca, Inc. in the United States, the paper proposes a tiered, source-anchored verification standard, keyed to source authoritativeness, procedural stage, and research-responsibility allocation within a legal team, capable of catching fabricated authorities before a judge relies on them rather than after an entire tier of appeal already has.