The Artificial Intelligence Committee of the Supreme Court of India released this preliminary draft to establish an institutional framework for the responsible adoption of AI across the Indian judicial system. Grounded in core principles-human primacy, judicial independence, transparency, accountability, data protection, and the rule of law-these regulations ensure AI serves strictly in an assistive capacity and never replaces human judgment.
Stakeholders and the general public are invited to submit views and suggestions via email to `office.regcc@sci.nic.in` by June 20, 2026
Permissible vs. Prohibited Uses
Permissible Assistive Uses (Regulation 19)
AI may be integrated into both administrative and specific assistive functions, including:
*Automated transcription of court proceedings (subject to verification).
*Translation of judgments, orders, pleadings, and legal documents.
*Legal research, precedent retrieval, citation verification, and document summarization.
*Filing assistance, defect scrutiny, and case resource allocation.
*Conversational AI assistants/guided chatbots to help litigants navigate court services.
*Accessibility services (e.g., text-to-speech, speech-to-text, Braille translation).
Prohibited Uses (Regulation 20)
The regulations enforce absolute prohibitions that cannot be relaxed or modified:
No Automated Adjudication: No judicial outcome (judgments, orders, findings of fact/law) can be reached solely through Algorithmic Decision-Making (ADM). Human authority remains determinative.
No Outcome Prediction: AI tools are strictly barred from predicting dispute outcomes.
No Automated Sentencing: AI cannot handle sentencing or adjudication independently; any output must remain advisory and subject to human review.
No Risk Scoring: AI cannot be used to assign scores estimating future behavior, flight risks, or recidivism.
Training Safeguards: Personal data cannot be used to train or test AI models without explicit authorized approval.
Oversight, Security, and Incident Management
To protect litigants and maintain system integrity, the draft mandates strict fallback and tracking measures:
Impact Assessments: Prior to approval, a comprehensive Technical and Ethical Impact Assessment evaluating system architecture, data source bias, and hallucination risks is required.
Controlled Testing: New software must undergo time-limited testing in an isolated, secure sandbox environment so it does not interfere with primary court operational networks.
AI Register & Incident Database: Every court must maintain an AI Register of deployed tools. Furthermore, an AI Incident Database will log malfunctions, errors, or data breaches. Malfunctions with potential legal consequences must be reported immediately for remedial action or tool suspension.
Mandatory Human-in-the-Loop (HITL): High-risk applications require human review, leaving final accountability and decision-making exclusively with human officers. Halting accountability by blaming an AI hallucination or "Black Box" opacity is explicitly forbidden.
Transparency, Privacy, and Private Vendors
Litigant Disclosure: Courts must inform parties if an AI tool materially assists in case management. Conversely, if a party or legal representative submits any AI-assisted pleading or evidence, they must formally disclose it via a designated certificate.
Data Protection: Systems must strictly comply with the Digital Personal Data Protection Act, 2023 and the IT Act, 2000. Principles of *Data Minimization* apply across system lifecycles. Sensitive judicial data cannot be transferred to an external system without explicit written authorization.
Private Vendor Restraints: Third-party providers need written authorization and are subject to periodic audits. Vendors must explicitly give the court a perpetual, royalty-free license for any customized tool; private entities are strictly banned from claiming exclusive intellectual property rights over tools developed using public judicial data or public resources. Models cannot be retrained or modified using court data without approval.
