By Anuradha Gandhi and Nihit Nagpal
Introduction
On June 3, 2026, the Supreme Court of India’s AI Committee released preliminary draft of the Regulations for Use of Artificial Intelligence (AI) in Courts, 2026 (“The Draft”). This marks a shift from scattered experiments with AI tools for translation, transcription, research and case management. With over 93,000 pending matters as of March 2026, the guidelines aim to ease processes for officials and citizens.[1] AI could help reduce backlogs through automated cause-list preparation, transcription, research, and defect scrutiny, but its use also raises serious concerns.
Documented Perils of AI in legal field
- Hallucinations-In Gummadi Usha Rani v. Sure Mallikarjuna Rao,[2] a trial court relied on fabricated precedents generated through AI, and the Supreme Court declared unequivocally that such reliance was “not an error in the decision making” but “misconduct” with legal consequences.
- Erosion of trust- courts rely on authenticity and precedent. If AI-generated content infiltrates judgments or submissions, public trust in judiciary could collapse
- Bias and discrimination– AI systems trained on skewed datasets may reinforce existing social or caste biases, leading to unfair outcome in sensitive cases.
- Accountability gap– when AI tools are used in drafting orders or assisting judges, it becomes unclear who is responsible for errors-the Judge, developer or the AI itself.
- Transparency issues-AI works on the Blackbox principle where the algorithm is a protected document
- Liability Gaps – If the system malfunctions the developer as well as the deployer can be held liable
- Data Privacy Risks- feeding sensitive cases details into AI systems could expose litigants’ personal information, especially if tools are not locally regulated or securely hosted.
Existing unstructured landscape governing usage and application of ai
Kerala[3], Bombay, and Gujarat[4] High Courts have each issued AI-related circulars and practice directions in 2025, though none amounts to a comprehensive standalone policy. Kerala takes the most conservative stance prohibiting generative AI in judgment drafting and mandating human countersignature on all AI outputs, with a data residency requirement. Bombay HC has focused on operational adoption, institutionalizing AI transcription of hearings and deploying a litigant-facing chatbot. Gujarat HC has leaned into efficiency, deploying predictive docket management and AI-assisted ADR routing, while remaining almost entirely silent on criminal proceedings where liberty interests are highest.
Innovation before Restraint and Presumption in favor of responsible AI adoption
While High Courts regulate use, the Supreme Court regulates principles, anchoring non-discrimination, explainability, and data governance in Articles 14 and 21.
Architecture of the Draft
The Draft is an ambitious document consisting ten chapters, fifty-seven regulations, and a detailed institutional architecture. The regulations apply to Supreme Court, High Courts, subordinate courts, tribunals and statutory commissions performing adjudicatory functions.
Chapter II: General Principles
Chapter II sets out fourteen general principles. Regulation 4 reserves the “ultimate authority to determine matters of law, fact and justice” exclusively to judicial officers. AI system shall be used solely in assistive capacity and shall be in compliance with the Rule of law (Regulation 5) protecting the rights of the individual. Every AI system shall be developed and deployed to meet high standards of transparency and explainability. The AI system incapable of explanation shall be restricted in high-risk applications.
Regulation 8 states that judicial officers cannot escape accountability by noting AI system’s hallucination, or opaqueness of Black Box.
Chapter III: Permissible and Prohibited uses
Unlike several foreign frameworks that regulate AI according to levels of risk such as European Union Artificial Intelligence Act (EU AI Act), the Draft creates two categories of permissible and non-permissible uses of AI. Regulation 19 enumerates a non-exhaustive list of permissible uses, all subject to prior written approval of the “Appropriate Authority” (the Apex Body at the Supreme Court level, and the AI Committee at the High Court level). Some of the permissible uses comprise case management, automated transcription, translation, legal research, conversational AI Assistant and guided chatbots, anonymization of documents, accessibility services, all of which require human verification.[5] Regulation 20 sets out an absolute list of prohibited uses, immune even from the Chief Justice’s relaxation power in Regulation 56. The non-permissible uses consist of not using Personal Information for training, testing or refining any AI, bail risk scoring, behavioral profiling, adjudication, undisclosed AI in any proceeding affecting personal liberty.[6]
Chapter IV: Institutional Framework
At the apex sits a permanent “Apex Body” at the Supreme Court, comprising judicial, technical and government members.Five standing committees report to the Apex Body: a Judicial Committee, a Technical Committee, a Committee on Infrastructure and Finance, a Case and Data Management Committee, and a Cyber Security Committee.
The Judicial Committee formed under Regulation 26 shall have the responsibility to innovate, develop and deploy AI tools to assist in adjudication in order to meet operational requirement of Courts and address other specific needs or concerns relation to adoption and use of AI in Courts.
Beneath the Apex Body, every High Court and the Supreme Court constitutes its own AI Committee (Regulation 33). Each AI Committee is supported by a dedicated AI Secretariat (Regulation 34). The Secretariat maintains the AI Register, the AI Incident Database, and has power to grant expedited approval for low-risk administrative tools.
Chapter V: Oversight, Audit and Incident Management
Every AI system requires a Technical and Ethical Impact Assessment before approval (Regulation 35). Periodic technical, legal, and ethical audits must be conducted at intervals not exceeding one year (Regulation 38). Crucially, for all Court AI Tools deployed within its jurisdiction audit are to be conducted exclusively in-house: “under no circumstances the source code, algorithms, datasets, or other architectural information shall be shared with any third party or private entity for an audit outside the Court premises” (Regulation 38(2)).
Concerns from the Regulations
The Disclosure Problem
Regulation 7 on transparency and explainability limits it to functioning, data input and decision logic. The transparency shall also include informing its limitations (including its margin of error), its capabilities, and the purpose of the systems along with providing the people that may be affected by such decision to be made aware of use of AI.
Regulation 40 grants supervising officer with full discretion to accept, modify or reject and AI-generated recommendation or output without requiring any transparency, standards or accountability. This can lead to arbitrary decision and creates accountability gaps.
Implementation of Regulation 38 Problem
Regulation 38 audit requirements is a significant safeguard, its effective implementation poses challenges.
A February 2026 report by DAKHS and Digital Future Lab commissioned by UNDP (“The DAKSH Report”) found that AI adoption in Indian courts had been largely ad hoc and poorly documented,[7] driven by individual champions rather than institutional policy. In this context, concerns lie regarding technical capacity of courts to conduct specialized audits on a recurring basis, particularly where AI systems involve complex machine learning or generative AI models.
Further, Uniform regulations for the use of AI across all courts and tribunals in India are far-fetched, as the country’s diverse literacy levels mean that one‑size‑fits‑all guidelines risk uneven implementation, misuse, and widening systemic gaps. India should adopt a phased, tiered framework that scales safeguards with local capacity and digital literacy so that innovation does not outpace fairness or accountability.
International Comparative Analysis
India joins global efforts to regulate judicial AI, with human centric approach identifiable across jurisdiction.
The Singapore Court’s Guide on the Use of Generative Artificial Intelligence Tools by Court Users does not require parties to make a pre-emptive declaration whenever AI use.[8] Instead, users must be prepared to identify AI-generated portions and explain their verifications steps if the court requests. This is a risk targeted model that differentiates obligations for different tasks that India’s draft should consider incorporating.
The European Union has taken a prescriptive position. Under Annex III, point 8 of the EU AI Act, systems deployed in the administration of justice to assist with researching or interpreting facts and law are classified as “high risk.” Thereby triggering obligations under Article 9 through 15 around human oversight and safeguards against automation bias. Instead of binary approach like India prescribing what is permissible and what is not, it instead classifies safeguards based on risk presented.
The United Kingdom also takes a principle based approach in terms of holding judicial officers personally liable for all content produced under their name, irrespective of AI use.[9] The guide titled “Artificial Intelligence Guidance for Judicial Office Holders” also states that information entered into public AI system is to be treated as being “published to the whole world”.
Operational since 2018, Brazil’s project Victor,[10] deployed at the Supreme Federal Court uses AI to identify whether appeals raise questions of sufficient “general repercussion,” thereby filtering cases before they reach full bench consideration.
AI for administrative purpose and case classification will offer a constitutionally safe entry point for judicial AI Integration.
China’s policy of Smart Courts by far is the most visionary approach. Since a 2017 national AI development plan, the Supreme Court of China built a national judicial infrastructure drawing on massive high quality judicial data. Despite such active integration of AI in judicial processes, China, like India maintains a “people focused” ideology and retains human responsibility for final judicial determinations.[11]
UNESCO also published guidelines in 2025 built around 15 universal principles to guide the adoption, governance and oversight of AI in the Judiciary along with the global toolkit on AI and the Rule of Law for the Judiciary.
Conclusion
The draft regulations marks a crucial step to formally integrate and regulate the use of AI in judicial processes. While earlier initiatives were largely court specific policies by the Hon’ble Kerala and Gujarat High Court, the present Draft serves as a uniform framework for binding all judicial bodies. Apart from a mere principle based framework, it also introduces institutional mechanism for oversight, transparency and accountability.
At the same time, certain provisions may benefit from further refinement. The mandatory disclosure requirement under Regulation 43 appears broad enough to capture routine and low risk uses of AI, while the audit framework under Regulation 38 will require substantial technical and institutional capacity to be effective in practice. Comparative experience from jurisdictions such as Singapore and the EU suggest that a more risk sensitive approach may serve India better.
Any AI system trained on historical judicial data in India will inherit the biases embedded in that data including systemic disadvantages faced by women, Scheduled Castes and Tribes, persons with disabilities, and speakers of languages other than Hindi and English, who are underrepresented in higher court judgments. India’s regulatory framework should require explicit bias audits for any AI system deployed in judicial contexts, and should establish standards for linguistic accessibility to ensure that AI tools designed to improve access to justice do not in practice exacerbate existing inequities.
Abhishekta Sharma, Junior Associate Advocate and Ishani Garg, Former Intern at S.S. Rana & Co., has assisted in the research of this Article.
[1] Namrata Banerjee, “The Master and the Machine”, Supreme Court Observer (16 April 2026) available at https://www.scobserver.in/journal/march-2026-pendency-climbs-to-over-93000-cases/
[2] SLP (C) No. 7575 of 2026, decided on 27 February 2026
[3] Kerala_HC_AI_Guidelines.pdf
[4] Gujarat_HC_AI_Policy – Merged.docx – Google Docs
[5] Regulation 19
[6] Regulation 20(1)
[7] DAKSH & Digital Futures Lab (UNDP), AI for Justice: Ethical, Fair and Robust Adoption in India’s Courts (February 2026)
[8] Registrar’s Circular No.1 of 2024, Guide on the Use of Generative Artificial Intelligence Tools by Court Users (Singapore Courts), effective 1 October 2024.
[9] Courts and tribunals judiciary, “Artificial Intelligence (AI) Guidance for Judicial Office Holders” (31 October 2025)
[10] International Bar Association, “AI-generated Evidence: The Brazilian Landscape”
[11] International Journal for Court Administration, “Lessons from China’s Smart Court Reform?”, available at https://iacajournal.org/articles/10.36745/ijca.679