India’s Supreme Court AI Rules Bar Algorithms From the Bench

India’s Supreme Court on June 4 released the “Regulations for Use of Artificial Intelligence (AI) in Courts, 2026,” the first comprehensive judicial AI governance framework produced by any major democracy. Prepared by the court’s AI Committee under Chief Justice Surya Kant, the draft covers every tier of the Indian judiciary from the apex court down to district tribunals and statutory adjudicatory bodies. The system collectively carries 55.8 million pending cases. Public feedback is accepted through June 20.

Beyond the headline governance rules, the draft builds a vendor clearance structure. No private technology company may participate in any court AI system without prior approval from a proposed permanent national apex body, and where court data helps build or improve an AI tool, the courts claim ownership or a perpetual royalty-free licence over its outputs.

55.8 Million Reasons to Move

• 55.8 million pending cases across all Indian court levels as of March 2026, per the National Judicial Data Grid (NJDG)
• 49 million of those in district courts alone, where most citizens first encounter the legal system
• 21 judges per million population, against a Law Commission recommendation of 50 per million
• 4,000+ courts already using AI-enabled transcription tools without any governing framework

India is writing judicial AI policy after the technology has already arrived. The Supreme Court’s SUVAS (Supreme Court Vidhik Anuvaad Software), a court-built translation tool, has converted judgments into 18 regional languages for years. AI-enabled transcription tools like Adalat AI, which already records real-time court proceedings across more than 4,000 state courts, ran without any governing framework until this month. The June 4 draft is the first attempt to set one across the entire system.

A 2018 Niti Aayog strategy paper calculated that, at then-prevailing disposal rates, clearing the backlog would take more than 324 years. Pending cases have nearly doubled since that paper, from 29 million to 55.8 million. More than 17 million of the current cases have been waiting over five years; over 180,000 have dragged for more than 30 years. Courts currently dispose of roughly 94 cases for every 100 new ones filed, per the India Justice Report’s 2025 analysis of high-court clearance rates, so the pile grows regardless of annual output. The drag on economic activity is measurable: judicial delays cost India more than 2 percent of GDP in foregone growth, according to research on the commercial-dispute backlog, because businesses settle well below fair value rather than wait out a decade-long trial.

Which Courtroom Tasks Get Approved

The Green List

The draft builds in what it calls a “presumption in favour of responsible AI adoption”; courts are encouraged to actively explore technologies capable of reducing delays or improving access to justice, provided those tools pass prior-approval processes and remain subject to human verification. The list of explicitly permitted uses runs the length of the administrative stack.

The Hard Bans and the COMPAS Shadow

The regulations state the governing principle explicitly, labeling it human primacy: AI must remain strictly subservient to judicial judgment, and the power to determine questions of law, fact, and justice vests exclusively with judges. The prohibition on bail assessment and recidivism scoring carries a specific international reference point. COMPAS (Correctional Offender Management Profiling for Alternative Sanctions), a proprietary risk-assessment algorithm widely used in US state courts, was the subject of a 2016 ProPublica investigation into US recidivism algorithms that found it classified Black defendants as high risk at nearly twice the rate of white defendants who did not reoffend. The controversy exposed a structural problem with any predictive criminal justice algorithm: the historical conviction data it trains on carries the biases of the system that generated it. India’s draft addresses that problem by prohibiting the category entirely.

Every AI System shall function solely in an assistive capacity and shall not supplant or compromise the independent exercise of judicial authority by a duly appointed judicial officer.

The draft then separately bars opaque systems, the kind whose internal reasoning cannot be explained, from any context affecting rights or personal liberty. A defendant who cannot understand how a system reached its output cannot meaningfully challenge it. The court encodes that asymmetry as a hard prohibition rather than a design guideline.

Draft Regulation 20 adds an institutional dimension: no AI system may be used for “the surveillance or continuous monitoring of judicial officers, advocates, litigants or any person connected with court proceedings.” An AI audit trail of judicial behaviour creates a form of institutional pressure on individual judges that the draft treats as incompatible with the independence the Constitution guarantees them.

The Fake Citations That Forced a Disclosure Rule

The disclosure requirements in the draft have a precise origin. In June 2023, a New York federal judge fined a lawyer $5,000 after a ChatGPT-generated brief submitted in Mata v. Avianca, Inc. cited cases that did not exist, the first known instance in a US federal court. By late 2025, researchers tracking AI hallucination sanctions in US courts counted the rate at two or three cases per day. Sanctions have followed in Massachusetts, California, Colorado, Texas, and Pennsylvania.

India’s Supreme Court was already tracking the problem at home. Before the draft was published, the court took judicial notice of AI-generated fake citations and sought responses from bar bodies on how to regulate them. The Kerala and Gujarat high courts had issued their own district-judiciary guidelines on AI use, creating a patchwork of state-level rules that the June 4 national framework supersedes.

The draft also proposes a dedicated AI Content Verification Authority to develop standards and protocols for verifying AI-generated outputs before they reach proceedings. That body would focus specifically on output accuracy, sitting alongside the Apex Body, which handles system-level approvals. The two-track structure reflects the underlying distinction: the hallucination problem (is this citation real?) and the governance problem (is this system approved?) are related but require different institutional responses.

Lawyers and litigants who use AI in preparing pleadings or submissions must file a prescribed declaration disclosing that use. Courts may then require further specifics: which AI system was used, the scope of AI assistance, and what verification steps were taken to check the output. If a pleading proves false because of AI-generated content, the filer bears full liability. There is no “the AI produced the error” defence available under the draft.

How the Governance Machine Is Built

The oversight structure the draft proposes is deliberately centralised and multi-tiered. A permanent Apex Body at the Supreme Court level would set national standards, approve AI systems for court use, and supervise implementation across every court in the country. Below it sit specialised committees covering judicial applications, technology, infrastructure, finance, cybersecurity, and data management. Judicial officers head each body; technology specialists function as support rather than as principals.

Each high court would maintain its own AI Committee and AI Secretariat. These bodies handle local approvals, investigate AI-related incidents, and conduct periodic reviews of deployed systems. Before any AI system goes live, it must pass two separate assessments: a technical evaluation and an ethical impact review that covers training data sources, bias risks, hallucination tendencies, cybersecurity vulnerabilities, and whether the system’s reasoning is explainable to the standard the black-box prohibition requires. Controlled-environment testing before large-scale deployment is also contemplated.

A Centre of Research and Excellence on Artificial Intelligence (CoRE-AI) would provide technical and legal support across the system, supplementing the Apex Body’s oversight function with continuous research capacity. Courts must maintain AI registers, incident databases, and annual transparency reports. High courts, tribunals, and statutory bodies performing adjudicatory functions would be required to publicly disclose which AI systems they run, what audits found, and what incidents were logged during each reporting period.

Compliance with India’s Digital Personal Data Protection Act (DPDPA), 2023, the country’s principal data privacy law, runs through all layers of the framework. Personal data cannot be used to train or improve any AI system without prior approval. Sensitive judicial information carries heightened protection standards. Vendors must contractually commit to data-use restrictions, cybersecurity obligations, audit rights, and direct liability for any harm caused by their systems inside the court ecosystem.

The Vendor Approval Gate

Prior Approval and Data Rights

The vendor provisions in the draft are commercially significant in ways the governance headlines don’t fully capture. Any private technology company seeking a role in court AI must obtain clearance from the Apex Body before deployment. The conditions are non-negotiable on one point: where a vendor builds or improves an AI system using court data or court resources, the courts retain ownership of that system or a perpetual royalty-free licence over its outputs. Courts are the data source and the client; vendors who want access to one must share rights over the other.

• Prior approval from the Supreme Court-level Apex Body is mandatory before any vendor engagement begins
• Contracts must specify court data ownership, data-use restrictions, and audit rights
• Vendors bear direct liability for harm caused by their deployed AI systems
• Ethical impact assessments covering bias, hallucination risk, and cybersecurity are required before deployment
• Opaque or unexplainable AI cannot be deployed in any matter affecting personal liberty
• Annual transparency reporting and AI audit trails are mandatory across all approved deployments

A Market in the Making

The approval requirement is also an entry ticket to a segment growing faster than most comparable sectors in India. The country’s legal AI market was valued at USD 29.5 million in 2024, per Grand View Research’s India legal AI market outlook, with the figure projected to reach USD 106.3 million by 2030 at a 23 percent compound annual growth rate. The broader Indian legaltech sector is expected to hit USD 1.28 billion in 2026, growing at 15.2 percent annually. India has more than 800 active legaltech companies, and sector funding in 2025 rose 781 percent compared to the prior year.

None of those figures account for what court-specific AI procurement at scale would add. The draft’s scope covers 25,000 sanctioned judges, hundreds of high courts, district courts, and tribunals, plus every statutory body with adjudicatory functions. The compliance requirements alone (ethical impact assessments, DPDPA-aligned data handling, mandatory audit trails) create a barrier that filters out vendors without existing legal-sector infrastructure in India. Companies already operating in court-adjacent workflows, whether transcription providers, legal research platforms, or case management software firms, are better positioned to meet those requirements than companies entering the judicial market from scratch.

Microsoft is already present through its Azure OpenAI partnership with SCC Online, which has built a conversational legal research assistant for Indian lawyers. Whether that deployment qualifies as a court system engagement under the draft’s definition will be among the interpretive questions the Apex Body faces once constituted.

Public feedback on the draft closes June 20. Until the Apex Body is formally constituted and the first AI systems are cleared for use, no court AI in India operates under this framework.