The Supreme Court of India’s AI Committee has released draft rules that would require lawyers to disclose when they use artificial intelligence tools in court filings, marking a significant step toward regulating AI across the nation’s judicial system. Titled the Draft Regulations for Use of Artificial Intelligence in Courts, 2026, the framework establishes a “human-in-the-loop” philosophy, allowing AI to assist with administrative and analytical tasks while reserving final judicial decisions exclusively for human officers.
Strict Disclosure Rules for Legal Practitioners
Under Regulation 43, lawyers who use AI to prepare documents, pleadings, or evidence must inform the court of the AI-assisted nature of that material at the time of submission. Judges may demand granular details about the specific AI system used, the extent of its contribution, and the exact verification steps taken to ensure accuracy. The rules target AI “hallucinations” – instances where large language models invent highly plausible but entirely fictional legal precedents, case laws, or statutory provisions – by taking aim at the practice of blaming AI opacity or system hallucinations as a defense for submitting harmful, inaccurate, or illegal content.
The draft explicitly prohibits advocates and judicial officers from blaming algorithmic opacity or system hallucinations as a defense. If a chatbot hallucinates a fake precedent, the signing lawyer will face the court’s wrath and potential disciplinary action. These measures aim to prevent the kind of algorithmic bias seen in Western jurisdictions, where predictive risk-scoring systems have reinforced socioeconomic disparities in bail and sentencing decisions.
The regulations distinguish between permitted and prohibited AI functions. Allowed uses include legal research, citation verification, document summarization, real-time translation, case management, scheduling, defect scrutiny, accessibility tools, and public-facing litigant chatbots. Prohibited applications encompass deciding cases, determining bail eligibility, assessing witness credibility, predicting litigant behavior, conducting judicial surveillance, and interfacing with internal deliberations.
Governance Structure and Vendor Controls
To oversee implementation, the Supreme Court proposes a permanent national regulatory body at the apex court level, staffed with judges, cybersecurity experts, technology specialists, and tech-law advocates. This apex body would set national standards and approve AI software. Each High Court would host operational AI committees and secretariats for compliance monitoring and enforcement.
A dedicated Centre of Research and Excellence on Artificial Intelligence (CoRE-AI) would evaluate emerging tools and provide technical support. All AI systems deployed in courts must undergo mandatory technical, legal, and ethical audits annually. High Courts will also be required to maintain public AI Registers and contribute to a centralized AI Incident Database to track malfunctions, data breaches, and algorithmic bias.
The framework also restricts private technology vendors, requiring explicit institutional approval for any AI deployment in Indian courts. Contracts must include safeguards ensuring that tools trained on Indian court data remain judiciary property or grant courts perpetual, royalty-free licenses. Data usage must comply with the Digital Personal Data Protection Act, 2023, mandating strict anonymization and minimization protocols.
Despite these safeguards, the draft codifies a “presumption in favor of responsible AI adoption,” encouraging courts to view AI as a catalyst for reducing systemic delays and expanding access to justice. Courts that reject approved AI tools must record written justifications for their decisions. Stakeholders have until June 20 to submit comments before the regulations are finalized and applied uniformly across all judicial tiers.
Technical Audit Mandates and Compliance Tracking
High Courts are required to maintain public AI registers documenting all approved tools, their vendors, and audit results. These registers must include system specifications, training data sources, and known limitations.
Vendor Approval and Data Ownership Safeguards
Private entities cannot deploy AI systems in Indian courts without explicit institutional approval. The apex body reviews each vendor’s security protocols, source code disclosures, and compliance history before granting clearance. Where ownership transfers are impractical, courts receive perpetual, royalty-free licenses. These provisions protect public investment in judicial datasets from commercial exploitation. Vendors must implement strict anonymization techniques and limit data collection to essential operational needs. Breach notifications must reach the apex body within seventy-two hours of detection.
