Supreme Court’s AI in Courts Regulations 2026: CLAT Guide

CURRENT AFFAIRS | JULY 13, 2026

Artificial intelligence has quietly entered India’s courtrooms — for translation, transcription, research and case management — and the Supreme Court has now decided that this entry needs rules before it deepens. On 3 June 2026, the Court’s AI Committee released the Draft Regulations for Use of Artificial Intelligence in Courts, 2026, a document that tries to answer a question every modern judiciary is wrestling with: how do you harness AI’s efficiency without letting an algorithm quietly take over the act of judging?

What the draft does

The draft regulations were released on 3 June 2026 and were open for public comments until 20 June 2026. Their scope is deliberately wide: they apply to the Supreme Court, all High Courts, and every court, tribunal or statutory commission performing adjudicatory functions. Crucially, they are not automatically binding — they come into force only on dates notified by the Chief Justice of India and the respective High Court Chief Justices, a design that keeps the judiciary in control of its own rollout.

At the heart of the framework are five core principles: human primacy, transparency, accountability, data protection and judicial independence. Everything else in the draft — what AI may do, what it may never do, and the institutions built to police the line — flows from these five commitments.

The absolute prohibitions

The most striking feature of the draft is its list of things AI is categorically barred from doing. These are described as “absolute and non-derogable” prohibitions. AI may not: decide cases; determine bail eligibility; “risk-score” a person to predict recidivism; assess the credibility of witnesses or parties; influence judicial deliberations; or monitor judicial officers, litigants or advocates — unless a law expressly authorises it.

The logic is that each of these tasks lies at the core of the judicial function, where human judgment, conscience and accountability cannot be delegated to a statistical model. The bar on recidivism “risk-scoring” is especially pointed, echoing the international controversy over tools like COMPAS in the United States — the subject of State v. Loomis — where an opaque algorithm’s risk score influenced sentencing and raised due-process alarms.

What AI is allowed to do

The draft is not anti-technology. It expressly permits a broad set of assistive uses — always with human control. AI may be used for legal research, drafting assistance, transcription of proceedings, translation of judgments, citation verification, and case, record and administrative management. These are the tasks where AI multiplies a judge’s or clerk’s productivity without touching the adjudicatory core.

But the permission comes wrapped in transparency duties. Litigants must be informed when AI has materially assisted in their matter. Lawyers must disclose AI use in their pleadings. And vendors face hard limits: they cannot claim exclusive intellectual-property rights over court data, and they cannot fine-tune their models on court data without written approval. This last set of rules is a quiet but powerful safeguard against the privatisation of the public record of justice.

The institutional architecture

To make the principles operational, the draft builds a four-tier structure. At the apex sits an Apex Body at the Supreme Court level. Each High Court gets an AI Committee. Each court gets an AI Secretariat for day-to-day implementation. And overseeing evaluation and research is the Centre of Research and Excellence on AI (CoRE-AI), established under Regulation 32.

CoRE-AI is the intellectual engine of the system. Composed of judges, lawyers, technologists and academics, it evaluates AI tools before they are deployed, maintains a central record of approved AI tools, and publishes white papers to guide the wider legal system. In effect, no AI tool reaches a courtroom without passing through a body that blends legal and technical expertise — a deliberate attempt to prevent unvetted software from shaping outcomes.

Guarding the record: data, vendors and IP

One of the quietly radical parts of the draft is how it treats the court’s own data. Court records are a public asset — the accumulated reasoning of the judiciary — and the regulations refuse to let private vendors capture them. A company supplying an AI transcription or research tool cannot claim exclusive intellectual-property rights over court data, and it cannot fine-tune its models on that data without written approval. This prevents a scenario in which a handful of legal-tech firms build proprietary advantages on the back of India’s public judicial record, and it keeps the courts, rather than their contractors, in control of how the record is used.

Data protection, one of the five core principles, ties directly into the Digital Personal Data Protection (DPDP) Act, 2023 and the privacy jurisprudence of Puttaswamy. Litigants’ personal information flows through pleadings, evidence and orders; an AI system trained or operated carelessly could leak or misuse it. By embedding data protection as a non-negotiable principle and layering vendor restrictions on top, the draft treats privacy not as an afterthought but as a design constraint on every tool that touches a courtroom.

Where India sits in the global picture

India is not alone in confronting AI in adjudication. The barred category of recidivism “risk-scoring” is a direct lesson from abroad: in the United States, tools like COMPAS — litigated in State v. Loomis — showed how an opaque algorithm could shade a sentencing decision while resisting meaningful challenge, because its inner workings were a trade secret. The Indian draft’s response is to keep such tools out of the adjudicatory core entirely rather than attempt to regulate their outputs after the fact. That is a deliberately cautious, rights-protective posture, consistent with treating judicial independence and due process as parts of the basic structure that cannot be traded away for efficiency.

Why “human-in-the-loop” is the organising idea

Read together, the five principles and the prohibitions all converge on a single design philosophy: the human must always remain in the loop. AI can search, draft, translate and organise — but a human judge must decide, a human must be accountable, and a human must be able to explain the reasoning. This is the practical meaning of “human primacy,” and it is what distinguishes assistive AI from autonomous decision-making.

The framework also confronts the transparency problem head-on. Because many AI systems are “black boxes,” the draft’s insistence on disclosure — to litigants and in pleadings — is what makes accountability enforceable. You cannot challenge an influence you cannot see; requiring disclosure keeps AI’s role visible and therefore contestable.