Regulating Artificial Intelligence In Indian Judiciary: From Institutional Experimentation To A National Framework

Update: 2026-06-11 08:30 GMT

Click the Play button to listen to article

The Indian judiciary's experiments with digital technology began in earnest with the e-Courts Mission Mode Project,^[2] which consists of three phases. Phase I (2007-2015) focused on foundational infrastructure, while Phase II (2015-2023) saw system-wide digital maturity via the Case and Information System 3.0, and the setting up of the National Judicial Data Grid. Phase III (2023–present) explicitly focuses on AI, Machine Learning, Optical Character Recognition, and Natural Language Processing deployment for case management, legal research, and translation.^[3] For the Supreme Court, tools including SUPACE for AI-assisted research and summarisation, SUVAS for multilingual judgment translation across 19 languages, TERES for real-time transcription, and LegRAA for generative AI legal research have been progressively introduced.^[4]

Before the existence of this consolidated national framework, High Courts in different parts of the country responded to AI with diverging postures. The Kerala High Court took a proactive, governance-first stance, by issuing a formal AI use policy and mandating Adalat AI for witness transcription across its district courts.^[5] The Gujarat High Court adopted a markedly cautious approach in its own AI policy, by expressing considerable reservations and emphasising the risks of over-reliance and the imperative of preserving judicial integrity.^[6] These contrasting postures illustrated precisely the regulatory gap that the Supreme Court's recent instruments seek to close.

The Supreme Court, in November 2025, released the White Paper on AI and the Judiciary (“White Paper”).^[7] The White Paper framed AI deployment as a principled governance challenge rather than a purely technical one. It identified hallucination as a concrete, already-observed risk in India, citing a Karnataka trial court drafting incident, an Income Tax Appellate Tribunal order recalled for fictitious precedents, and a case filing containing ChatGPT-fabricated quotes.^[8] It flagged algorithmic bias, illustrating concerns through the US COMPAS controversy,^[9] the erosion of confidentiality through public AI tools, and the 'black-box' opacity problem undermining due process. Its core recommendations included: establishing AI Ethics Committees, prioritising secure in-house tools over public chatbots, maintaining disclosure and audit trails, mandatory independent verification of all AI outputs, and treating judicial officers as personally responsible for AI-assisted material produced in their name.^[10]

On the international front, the regulatory landscape concerning deployment of AI in judicial systems reflects a clear consensus: AI in courts must assist, never adjudicate. The chief multilateral instruments in this regard include the UNESCO's Recommendation on the Ethics of AI, 2021 (applicable to 194 member states),^[11] the OECD AI Principles, 2019 (updated in 2024),^[12] and the EU AI Act 2024 (which classifies the administration of justice as a 'high-risk' domain requiring the highest tier of regulatory compliance).^[13] These instruments collectively establish transparency, accountability, human oversight, and bias prevention as foundational non-negotiables.

The United Kingdom's AI Action Plan for Justice released in July 2025, set out the Government's approach for responsible and proportionate AI adoption across courts, tribunals, prisons, probation and supporting services.^[14] The United Kingdom later issued a detailed guidance in October 2025 requiring that judicial officers never enter confidential information into public AI tools, mandating accuracy checks before any reliance, and emphasising personal responsibility for all AI-assisted outputs.^[15]

In Brazil, the National Council of Justice has set out in relation to development, use and governance of AI solutions within the judiciary – the principles of respect for fundamental rights; due process; human oversight and risk based supervision; transparency, explainability, traceability and auditability; bias prevention; and data protection.^[16] Through its Sinapses platform, Brazil centralises supervision, control, and auditing of judicial AI nationwide.^[17] Canada's Judicial Council expressly prohibits delegation of decision-making authority to AI systems.^[18] Singapore's court user guide places full responsibility for AI-generated accuracy on the submitting party.^[19] Across all jurisdictions, the universal safeguard is a meaningful human-in-the-loop, not for a nominal rubber-stamp review, but genuine critical engagement.

Published on 3^rd June 2026, the Supreme Court of India's 'Draft Regulations for Use of Artificial Intelligence in Courts, 2026' (“Draft AI Regulations”)^[20] represent India's first attempt at a comprehensive, national, binding judicial AI governance framework (rather than an advisory guidance), applicable across the Supreme Court, all High Courts, and all tribunals and statutory commissions performing adjudicatory functions.^[21] Its core governance principles are as follows – AI must be strictly assistive and subservient to judicial authority; final authority on law, fact, and justice vests exclusively in judicial officers; AI must actively avoid discrimination; opaque and unexplainable systems face heightened scrutiny; and accountability for AI use rests personally on the officer using the tool.

All forms of permissible uses require prior written approval from the relevant nominated officers/ Appropriate Authority (as the case may be), and include - case management, transcription, translation, legal research and summarisation, and administrative analytics.^[22] Absolutely prohibited uses include algorithmic adjudication without mandatory human review, risk scoring for bail or recidivism, outcome prediction, AI surveillance of court users, and any use compromising the confidentiality of judicial deliberations.^[23] The Draft AI Regulations establish an 'Apex Body' at the Supreme Court, standing committees, and court-level AI Committees with dedicated AI Secretariats at every tier of the national judicial hierarchy.^[24]

The Draft AI Regulations are in clear intellectual continuity with the White Paper, operationalising its recommendations with binding legal force. Both are centred on human primacy, treat hallucination and fabricated citations as concrete risks, and require mandatory verification and disclosure. The Draft AI Regulations go further by establishing a detailed, non-derogable list of prohibited uses;^[25] mandating disclosure certificates from parties using AI in submissions;^[26] and creating a comprehensive national institutional architecture that builds much further upon the regulatory architecture envisioned in the White Paper.

However, four significant gaps require attention in the finalisation process. First, the restriction of audits to in-house processes, expressly prohibiting sharing of source code or architecture with third parties,^[27] runs counter to global consensus that external independent technical review is essential to public trust.^[28] A structured independent audit mechanism, operating under strict confidentiality obligations, should be incorporated.

Second, the litigant grievance mechanism, for those harmed by prohibited use of AI in judicial systems, is underdeveloped.^[29] The Draft AI Regulations do not specify the timelines, evidence standards, or compensation for individuals harmed by AI misuse, a gap that should be addressed by incorporating the UNESCO Global Toolkit on AI and the Rule of Law for the Judiciary's contestability safeguards,^[30] including rights to challenge algorithmic decisions and request human review.

Third, explainability standards are stated as a principle but not operationalised.^[31] Minimum technical standards, drawing on the US's National Institutes of Standards and Technology framework^[32] should be prescribed for systems used in rights-affecting contexts. These standards include – requirement for systems to provide reasons and evidence for outputs, explanations to be made understandable to users, and ensuring explanations reflect the actual process.

Fourth, binding training standards are absent.^[33] The training obligation should specify minimum content, frequency, and compliance accountability across the entire judicial system, to prevent the requirement from remaining aspirational in practice.

The Draft AI Regulations represent a creditable and comprehensive step forward. Their absolute prohibition on algorithmic adjudication and risk scoring is correctly and unambiguously framed. With targeted revision that address the identified lacunae, India's framework has the potential to become a leading model of judicial AI governance, one that other global judicial systems may in time seek to follow.

The Authors Varun Mehta & Ashirbad Nayak are Director and Senior Associate respectively in the Public Policy Practice of Cyril Amarchand Mangaldas. Views are personal.

Regulating Artificial Intelligence In Indian Judiciary: From Institutional Experimentation To A National Framework

Similar News

Women's Property Rights In India: Law On Paper Vs. Reality

Constitutional Pathology Of Delayed Adjudication: Reservation Without Preservation

Cross Border Insolvency In India: What Rules Must Say

Banking On Takeovers: RBI's New Acquisition Finance Regime In India

Coded Collusion: Algorithmic Pricing In Indian Aviation And Structural Failure Of Competition Law

Mihir Rajesh Shah : Did Supreme Court Create Two Classes Of Arrestees?

Rule Of Law As Economic Infrastructure: Why The Bar Remains Its Most Immediate Guardian

Leaked Chats, Media Trials And Growing Privacy Debate In India

The Devil Claims Jhumkas, And India Cannot Defend It! Why Traditional Cultural Expressions Need Serious Legal Attention

Advocates Act And Politicians Returning To The Bar

No Room For Bystanders In A Constitution That Outlawed Untouchability