AI May Translate the Judgment, Never Write It: The Supreme Court Draws Its Line
Draft Regulations for Use of AI in Courts, 2026 permit research and transcription but bar risk scoring, bail prediction and black-box systems in liberty matters
What happened
The interesting thing about this draft is not that it regulates AI but where it refuses to let AI go. Most technology-in-governance questions are answered with efficiency arguments; this one is answered with a constitutional boundary — assistive functions on one side, adjudicative judgement on the other. Learn the line and the reasoning behind it, because it is the template India will likely extend to every other domain where an algorithm might touch someone's liberty.
The Line: Assistive vs Adjudicative
What AI May and May Not Do in Court
| PERMITTED (assistive) | PROHIBITED (adjudicative) |
|---|---|
| Legal research, citation verification | Risk scoring |
| Transcription, translation | Recidivism prediction |
| Drafting assistance, summarisation | Bail evaluation |
| Case management, scheduling | Witness credibility assessment |
| Record management, accessibility | Black-box systems in liberty matters |
Source: Supreme Court of India, draft Regulations for Use of Artificial Intelligence in Courts, 2026
The Draft Regulations for Use of Artificial Intelligence in Courts, 2026 were released by the Supreme Court's AI Committee in June 2026 for public consultation, and would apply to the Supreme Court, High Courts, subordinate courts, tribunals and statutory adjudicatory bodies, with each court phasing adoption at its discretion.
●The framework rests on a functional separation. **Permitted:** legal research, citation verification, drafting assistance, translation, transcription, case management, scheduling, record management, document summarisation and accessibility support — with mandatory disclosure by lawyers and litigants of AI use in pleadings. **Prohibited:** risk scoring, recidivism prediction, bail evaluation and witness credibility assessment; black-box systems are barred outright in matters touching personal liberty.
●The declared principle is that the power to decide questions of law, fact and justice vests exclusively in judges.
●Institutionally, the draft creates a permanent Apex Body at Supreme Court level comprising Supreme Court and High Court judges, MeitY officials and finance and cybersecurity experts; AI Committees and secretariats in courts; five specialised committees; and a Centre of Research and Excellence on Artificial Intelligence (CoRE-AI) staffed by technology, law and academia experts to evaluate tools.
●Compliance obligations include Technical and Ethical Impact Assessments, annual audits, AI registers and incident databases, notification of system failures within 24 hours, and limits on private vendors retaining data or fine-tuning models on court material.
The line is functional, not technological: AI may do anything that prepares a decision and nothing that constitutes one.
◎ In Simple Words
Indian courts have a huge backlog of cases, and computers could help — by translating judgments into other languages, typing out what witnesses say, or finding earlier cases that are relevant. The Supreme Court has written draft rules saying courts may use AI for exactly those helping tasks. But it also says AI must never be used to decide anything about a person — not whether they get bail, not whether they are likely to commit another crime, not whether a witness is telling the truth. Those decisions must stay with a human judge. Lawyers also have to say when they have used AI.
Factual Pointers
Practice · 2 questions
With reference to the draft Regulations for Use of Artificial Intelligence in Courts, 2026, consider the following statements:
1. They permit the use of AI for transcription, translation and legal research.
2. They permit AI-generated risk scores to assist judges in bail decisions.
3. They require lawyers and litigants to disclose the use of AI in pleadings.
Which of the statements given above are correct?
The proposed 'CoRE-AI' under the draft AI regulations for Indian courts refers to:
Mains Practice Questions
"Artificial intelligence may prepare a judicial decision but may never constitute one." Examine the constitutional basis of this distinction with reference to the draft Regulations for Use of AI in Courts, 2026. (250 words, GS2)
Explainability is a legal requirement before it is a technical one. Discuss with reference to the prohibition on black-box systems in matters affecting personal liberty. (250 words, GS2)
Technology addresses the symptoms of judicial delay rather than its causes. Critically examine. (150 words, GS2)
Frequently Asked
· People also askWhat are the draft AI regulations for Indian courts?
The Draft Regulations for Use of Artificial Intelligence in Courts, 2026 were released by the Supreme Court's AI Committee in June 2026 for public consultation. They would apply to the Supreme Court, High Courts, subordinate courts, tribunals and statutory adjudicatory bodies, with each court phasing adoption at its discretion.
Prelims · GS2They are the first systematic framework for AI in Indian adjudication. Their organising principle is that the power to decide questions of law, fact and justice remains exclusively with judges.
SOURCE Supreme Court of India, notice dated 3 June 2026
What uses of AI do the draft regulations permit?
Legal research, citation verification, drafting assistance, translation, transcription, case management, scheduling, record management, document summarisation and accessibility support. Lawyers and litigants must disclose their use of AI in pleadings.
GS2 · JudiciaryAll of these are assistive functions that prepare a decision without constituting one — the functional boundary the framework is built around. Translation and transcription address genuine bottlenecks in a multilingual judiciary.
SOURCE Draft Regulations for Use of AI in Courts, 2026
What uses of AI are prohibited?
Risk scoring, recidivism prediction, bail evaluation and witness credibility assessment are expressly barred, and black-box systems are prohibited outright in matters affecting personal liberty. These are adjudicative functions reserved exclusively to judges.
GS2 · PolityThe prohibition on non-explainable systems reflects an Article 21 requirement: a person whose liberty is curtailed is entitled to reasons, and a system that cannot explain its output cannot supply them.
SOURCE Draft Regulations for Use of AI in Courts, 2026
What is CoRE-AI?
The Centre of Research and Excellence on Artificial Intelligence — an integrated body of experts in technology, law and academia, constituted by the proposed Apex Body, to provide research and legal guidance on AI in court processes and to evaluate tools before deployment.
PrelimsIt sits alongside a permanent Apex Body comprising Supreme Court and High Court judges with MeitY, finance and cybersecurity officials, and five specialised committees.
SOURCE Draft Regulations for Use of AI in Courts, 2026
Why must lawyers disclose AI use in pleadings?
Because generative systems produce plausible but non-existent case citations. Courts in several jurisdictions have sanctioned lawyers who filed fabricated references without verification. Disclosure places responsibility for accuracy explicitly on the filing advocate, and preserves the adversarial process by informing the other side.
GS2 · EthicsProfessional responsibility for the accuracy of a filing is not transferable to a tool — which is why the draft pairs disclosure with a mandatory citation-verification function.
SOURCE Draft Regulations for Use of AI in Courts, 2026
How does India's approach compare internationally?
India's draft prohibits adjudicative AI outright while permitting assistive uses. The European Union's AI Act instead classifies justice-administration AI as high-risk, imposing transparency and human-oversight duties rather than bans. The United States permitted risk-assessment tools in sentencing and litigated the consequences afterwards.
GS2 · ComparativeIn State v. Loomis (Wisconsin, 2016) the COMPAS recidivism algorithm was allowed with warnings about its proprietary methodology. India forecloses at the outset the problem American courts have had to manage after the fact.
SOURCE EU AI Act; State v. Loomis (2016)
Will AI reduce judicial pendency in India?
Only marginally. Translation and transcription are genuine bottlenecks that AI can ease, but the binding constraints on case disposal are judicial vacancies, adjournment culture and procedural delay — none of which AI addresses. Treating it as a solution to pendency risks substituting visible technology for harder institutional reform.
GS2 · GovernanceCapacity is also uneven: subordinate courts lacking basic digital infrastructure will struggle with impact assessments and audits, so discretionary phasing may mean the framework binds well-resourced courts and bypasses others.
SOURCE eCourts Mission Mode Project; National Judicial Data Grid