The AI Committee of the Supreme Court of India has just unveiled a colossal leap into the legal frontier: the draft Regulations for Use of Artificial Intelligence (AI) in Courts, 2026. The document, which was released on June 3, 2026, isn’t just about reducing the number of files or clearing up filelists, it’s a full-scale constitutional border wall built to prevent artificial intelligence from sucking the human justice system dry.
The judiciary is actively seeking public and stakeholder input until 20 June 2026, before the judiciary adopts these rules as policy at all levels of the legal system – from the Supreme Court down to the local tribunals.
The Core Dogma: Human Primacy
Ignore the legalese and the one goal of the framework is clear: human primacy.
The draft explicitly prohibits AI from making final decisions. Except that a machine can’t decide a case, can’t hand out a sentence, and can’t tell if a witness is lying on the stand. The draft calls for a tight “Human-in-the-Loop” design.
This principle is clearly stated in the text, which allows no judge or court officer to explain an incorrect order based on a software glitch or an AI “hallucination”. If a judge employs an AI summary tool and it gives him a distorted version of the truth, the judge is clearly at fault.
The Lawyer’s Burden and the No-Hallucination Rule
The days of using the law-drafting “plug and play” feature with ChatGPT are over for practitioners. There is a new and strict, mandatory disclosure regime in the framework.
The Disclosure Mandate: Draft Regulation 43(3) states that a lawyer or litigant who relies on an information assistance tool to draft any pleading, submission, document, or evidence must make a formal disclosure at the time such pleading, submission, document and/or evidence is filed.
The Court’s Counter-Query: The judges have the power to seek details of the exact AI system, how it was used, and what human checks were performed to validate the work.
No Algebraic Protections: Wrongly submitted AI-generated filings that contain fabricated citations, references or facts are subject to the full range of professional misconduct or contempt liability. The machine can not be blamed as a defence, as this is clearly prohibited by law.
Use and abuse of the device are prohibited.The use and abuse of the device is forbidden.
The draft starts to separate the efficient running of the back office from the reasoning of the front office, judges.
Where AI is Invited
The framework is open to automation of administrative and technical tasks. This comprises automatic case scheduling, court list management, legal research, citation checking, real-time transcription and translation of complicated court decisions into regional Indian languages. It also enables the use of AI-powered chatbots to aid ordinary citizens in navigating the complex corridors and filing desks of the court.
Where AI is Banned
Fierce red lines are drawn “absolute and non-negotiable. The draft banish entirely black-box risk software used in making bail decisions, monitoring recidivism or predictive profiling. It bans opaque AI systems that impact on individual liberty and prevents automated monitoring of judges, lawyers or litigants in the judicial system.
Data security and institutions’ accountability for tracking data.
One way to avoid this being a toothless paper policy is to establish a multi-layered, dense regulatory framework.
[Apex AI Body (Supreme Court Level)]
│
(Directs Policy)
▼
[AI Committees (High Court Level)]
│
(Oversees Implementation)
▼
[Dedicated AI Secretariats] ──► (Technical, Cyber & Data Audits)It is a hierarchical control process designed to test out software environments (or “sandboxes”) prior to the implementation of any piece of software on real court data. Each tool should undergo annual cybersecurity assessments and be required to have open AI incident logs while also being held strictly accountable for adhering to India’s Digital Personal Data Protection (DPDP) Act, 2023.
The framework explicitly prohibits the sending of sensitive information out to a third party, external, commercial database, which means that there is a heavy restriction on access by vendors to the ongoing legal files.
