Legal AI in India: AI research, court rules and citation verification
By Utsav Patel, Founder, Vakilat
Last updated 4 November 2026
The Indian legal profession is navigating its most significant statutory and technological transformation. Between 2024 and 2026, two shifts converged: the legislative replacement of colonial criminal codes with the Bharatiya Nyaya Sanhita (BNS), Bharatiya Nagarik Suraksha Sanhita (BNSS), and Bharatiya Sakshya Adhiniyam (BSA), alongside the rapid integration of generative artificial intelligence into legal research and litigation drafting.
While consumer chatbots offer speed, unverified adoption has triggered serious judicial strictures before the Supreme Court of India and High Courts over fabricated citations. This guide provides an authoritative reference for advocates, law firms, and judicial officers on using legal AI safely, understanding court directives, and applying deterministic verification before citing authorities in court.
The rise of legal AI in India
For over thirty years, computer-assisted legal research in India relied on rigid Boolean keyword queries (AND, OR, NOT, NEAR). If an advocate searching for the quashing of an FIR in a non-compoundable matrimonial dispute failed to query the exact words used by the bench, vital authorities were missed.
Domain-specific legal AI models have transformed this workflow through natural language understanding:
- Natural Language Inquiries: Counsel can ask: “Can anticipatory bail be granted after a charge-sheet is filed without taking the accused into custody?”
- Contextual Retrieval: The engine maps the inquiry directly to Section 438 CrPC and Section 482 BNSS, retrieving landmark rulings regardless of specific keyword phrasing.
- The Risk of Generic LLMs: General chatbots (like ChatGPT, Gemini, and Copilot) operate on probabilistic next-token prediction. They generate text that sounds authentic rather than retrieving official judicial records from gazettes, often inventing non-existent case names, bench compositions, and paragraph numbers.
Unique challenges of Indian law
Applying artificial intelligence to Indian law involves structural realities unmatched in other common law jurisdictions:
1. Three-Tier Court Hierarchy
Under Article 141 of the Constitution, Supreme Court declarations of law bind courts across India. Beneath it, 25 High Courts exercise constitutional and appellate supervision. Conflicting High Court interpretations create regional splits that generic AI models cannot resolve without explicit jurisdictional filtering.
2. Bilingual and Vernacular Pleadings
While High Courts and the Supreme Court conduct appellate proceedings in English under Article 348(1), trial evidence originates in regional languages: FIRs, panchnamas, witness statements under Section 161 CrPC (Section 180 BNSS), and local testimonies. Reliable legal AI must bridge vernacular trial records with English appellate jurisprudence without distorting facts.
3. The 1 July 2024 Statutory Overhaul
The BNS, BNSS, and BSA repealed the IPC, CrPC, and Evidence Act on 1 July 2024. Under Section 531(2)(a) BNSS and Section 358 BNS, offences committed before 1 July 2024 remain governed by the IPC, and pending investigations continue under the 1973 Code. Legal research tools must maintain bi-directional crosswalk indexing between old and new codes.
Judicial strictures on AI hallucinations
Between 2024 and 2026, Indian courts moved decisively from exploratory tolerance to formal disciplinary action against advocates submitting unverified, AI-generated citations:
Supreme Court: Gummadi Usha Rani v. Sure Mallikarjuna Rao (SLP (C) No. 7575/2026)
A trial court disposed of commissioner report objections by relying on four fabricated case laws generated by an AI chatbot. Setting aside the order, the Supreme Court (Vikram Nath and Satish Chandra Sharma, JJ.) held:
“A decision based on such non-existent and fake alleged judgments is not an error in the decision making. It would be a misconduct and legal consequence shall follow.”
Supreme Court: Pooja Ramesh Singh v. J&K Bank Ltd. (2026 INSC 668, 2026 SCC OnLine SC 1258)
The NCLT admitted a Section 7 insolvency petition relying on automated AI research that invented fake judicial authorities, which the NCLAT failed to catch. Setting aside both orders, the Supreme Court ruled:
“It is a misconduct on the part of an advocate to cite such judgments without verification.”
Supreme Court: Vijay Ghanshyam Gadiya v. Union of India (2026 INSC 947)
Quashing a Rs 425 crore penalty based on non-existent authorities, the Supreme Court clarified the role of artificial intelligence:
“Assistance can never be substituted for adjudication. AI may well serve as training wheels but entrusting it with the pilot's seat would be both imprudent and dangerous.”
Supreme Court Draft Regulations for Use of AI in Courts, 2026
Published by the Supreme Court AI Committee, Regulation 43(3) requires advocates using AI to prepare pleadings or submissions to file a declaration in Annexure I. Under Regulation 43(6), claiming an error was caused by an AI system is expressly barred as a defense.
How Vakilat solves citation risk
Generic AI tools generate text probabilistically. Vakilat replaces probabilistic guessing with a deterministic verification pipeline built specifically for Indian litigation:
- Official Gazette Grounding: Answers are retrieved from the primary digital corpus of Supreme Court judgments (1950 to present), aligned with Supreme Court Reports (SCR) and official Neutral Citations (INSC).
- Word-for-Word Quotation Checks: Every quotation is tested against the official court record by character-level string verification. If a single word is altered, quotation marks are withheld and an audit note is issued.
- Exact Paragraph Linking: When the bench asks: “Which paragraph, counsel?”, Vakilat links each authority directly to the numbered paragraph, allowing advocates to verify the source before citing it.
- Dual-Model Adversarial Verification: One model synthesizes legal research; a second independent model verifies each proposition strictly against the primary passages, flagging any extrapolation.
- Statutory Crosswalk Engine: Bi-directional mapping between old and new codes (IPC to BNS, CrPC to BNSS, Evidence Act to BSA) with date-of-offence validation under Section 531 BNSS.
Six checks before you cite
Before submitting any judicial precedent in a petition, written submission, or oral argument, verify these six checkpoints:
Overruled and Per Incuriam Check
State of U.P. v. Synthetics and Chemicals Ltd. (1991) 4 SCC 139
Confirm that the judgment has not been overruled, modified, or reversed by a larger bench, and that it was not delivered per incuriam in ignorance of an applicable statute or binding precedent.
Article 141 Bench Strength Hierarchy
Central Board of Dawoodi Bohra Community (2005) 2 SCC 673
Under Article 141 of the Constitution, a decision of a 5-judge Constitution Bench prevails over a 3-judge bench, and a 3-judge bench prevails over a 2-judge bench. Check the exact bench composition.
Ratio Decidendi vs. Obiter Dicta
State of Haryana v. Ranbir (2006) 5 SCC 167
Isolate the core legal principle necessary for deciding the dispute (ratio decidendi). Observations made by the way (obiter dicta) carry persuasive weight but are not binding law under Article 141.
Paragraph Accuracy and Official Neutral Citations
Supreme Court Neutral Citation Circular (2023 INSC)
When the bench asks: 'Which paragraph, counsel?', have the exact paragraph ready. Never cite headnotes or summaries without verifying the primary court gazette text word-for-word.
Factual Parity Verification
Padmasundara Rao v. State of Tamil Nadu (2002) 3 SCC 533
Precedents cannot be read like statutes. A judgment is an authority only for what it actually decides on its specific facts. Confirm factual parity before relying on general judicial observations.
Statutory Transition Check (BNS, BNSS, BSA vs IPC, CrPC, IEA)
Section 531(2)(a) BNSS and Section 358 BNS
Offences committed prior to 1 July 2024 remain governed by the IPC, with pending trials governed by CrPC. Offences committed on or after 1 July 2024 fall strictly under BNS and BNSS.
Chamber standard operating procedures
To integrate legal AI safely, chambers should implement these practical operating standards:
- Mandatory Verification Sign-off: Maintain an internal sign-off record for every draft. An associate or counsel must verify each citation on the official e-SCR or High Court portal before filing.
- Compliance with Court AI Declarations: Where court rules require AI declarations (such as Regulation 43 of the Draft AI Regulations), ensure declarations accurately reflect human verification.
- Client Confidentiality: Never paste unredacted client files, commercial trade secrets, or privileged briefs into general consumer cloud AI tools. Use legal AI infrastructure like Vakilat where inputs are protected and never used to train external models.