Introduction
In February 2026, during the hearing of a Public Interest litigation, Justice B.V. Nagarathan encountered a case named Mercy v. Mankind, a case cited before the Supreme Court that was entirely non-existent. This was not an isolated incident and reflects a novel and acute problem that the judicial system is bogged down in. For instance, in March, a trial court had passed a judgment relying on judicial precedents generated by hallucinations, using Artificial Intelligence. Turning its attention to such invidious use of AI, the court directed that it would be tantamount to misconduct.
The issue had surfaced even earlier. In September 2025, a petition before the Delhi High Court was withdrawn after the respondents demonstrated reliance on fabricated authorities. A judgment that consisted only of 27 paragraphs, the purported non-existent paragraphs 73 and 74 were cited. Similarly, in January 2026, the Bombay High Court took judicial concerns over AI-generated legal filings a step further by imposing costs of Rs. 50,000 on a litigant whose written submission bore “give-away features” of AI-generated content. Justice M. M. Sathaye criticised the practice of filing unverified AI-generated material, warning that such submissions waste judicial time and hinder the efficient administration of justice.
On June 3, 2026, the Supreme Court of India’s Artificial Intelligence Committee released the draft version of its Regulations for Use of Artificial Intelligence in Courts, 2026 (“Draft Regulations”). This document constitutes an attempt on the part of the Supreme Court to regulate the use of AI in one of the most constitutionally significant domains. The Indian judiciary handles hundreds of millions of cases each year through its Supreme Court (“SC”), twenty-five High Courts and thousands of other lower courts and tribunals. The use of AI is not a theoretical construct in this context; several technologies used for transcription, cause-list preparation and legal research are being actively deployed. The Draft Regulations thus constitute a regulatory response to the judiciary’s ongoing adoption of artificial intelligence, rather than a pre-emptive framework for its future use.
The Architecture and Philosophical Foundation of the Draft Regulations
The structure of the draft runs across ten chapters, moving from definitions and scope through governing principles and permitted and prohibited uses to institutional grievances. Moreover, it also includes definitions for terms such as “hallucination,” “large language model,” and “black box” that suggest a real understanding of the way these systems operate and not some copy-paste version of a foreign rulebook.
It follows one simple rule: AI can be used to help, but never to judge. Thus, the independent judicial authority will never be undermined by AI. This is done by the requirement of Human-in-the-Loop elements when there is a risk to an individual’s liberty, by ensuring that there is an absolute prohibition on algorithmic adjudications, and by ensuring accountability elements, which hold a liability for an officer whose following directive is based on AI output, regardless of what the output may have been. The second principle prohibits any AI that reinforces any bias based on race, caste, religion, sex, gender, disability, language, or economic status, which directly correlates to Articles 14 and 15 of the Indian Constitution.
Substantive Prohibitions and Permitted Uses
The key provisions of the draft are those contained in Regulation 20, which specifies an exhaustive list of absolute (unsurpassable) bans or substantive prohibitions, and what is remarkable about these bans is that they embody a theory of judicial integrity. The bar is set at the lowest level of prohibition against algorithmic adjudication and sentencing without human intervention, and all comparative approaches reach that far. What extends that is the prohibition against any kind of risk scoring (e.g., for flight risk, likelihood of re-offending, who should be granted bail, and witness credibility); the prohibition against profiling parties or predicting their future behavior of litigants expands this argument beyond the courtroom to the litigants, holding that using AI to build up behavioral profiles of litigants before courts is fundamentally incompatible with the right to be heard on their terms. The bar on electronic surveillance of judges, advocates, and litigants protects the three parties of the courtroom from any type of ambient surveillance that AI systems naturally generate when deployed, and the prohibition on introducing AI-generated output as evidence without first disclosing its source formalises what should have been an obvious prohibition, but what obviously was not in the events that led up to this draft.
This is inclusive of a ban on employing personal data in enhancing an Artificial Intelligence System without the preliminary requirement of an individual’s data protection-compliant consent, a measure which prioritises a person’s information autonomy over the system’s data-hungry nature even in a state institution.
On the other hand, there exists a list of permitted uses contained in Regulation 19 that focuses on being illustrative and not exhaustive: case management, required human-supervised transcription, translation, legal research and citation verification, verification of document authenticity, and anonymising judgments. There is a structural point of contrast with Regulation 20. The permitted list is open and subject to the approval process, whereas the prohibited list is closed and has no exceptions, be it by court or by whomever. However, it is not a mistake as many may think; rather, it is a design decision that signals that the draft is not to halt the progress towards adoption of AI but to guide it. At first glance, it appears to be a framework that is more “relaxed,” more “open,” and more “liberal” than it claims; more “relaxed” in terms of process, in terms of administration, and in terms of access; and more “open” and more “liberal” in terms of adjudication, prediction, and surveillance.
The Enforcement Architecture
The enforcement architecture of India’s draft regulations operates through a structured, multi-layered hierarchy designed to ensure accountability, auditability, and human oversight. This includes the creation of a permanent Apex Body located at the Supreme Court headed by the Supreme Court Judge and composed of members from High Court Chief Justices; a Ministry of Electronics and Information Technology official; technical and cybersecurity experts approving AI systems and publishing annual governance reports; an AI committee in each High Court designated as the “Appropriate Authority” for that jurisdiction to supervise local implementation and compliance; developing an AI Secretariat with a District Judge (or equivalent) as its head to conduct research and evaluate AI, and establishing a Centre for Research & Excellence on Artificial Intelligence (CoRE-AI) to continuously conduct research and evaluate tools on an ongoing basis. Every AI system must clear a rigorous Technical and Ethical Impact Assessment before its deployment, checking for bias, hallucination risks, and data vulnerabilities (such as under the Digital Personal Data Protection Act, 2023. AI systems must undertake annual ethical and legal audits after distribution on these same criteria, as well as additional aspects of performance (e.g., quality of data used to train the AI System). The courts are also obligated to maintain public AI registers and an AI Incident Database for tracking any algorithm malfunctions, errors or breaches/
Beyond the judiciary, the Ministry of Electronics and Information Technology (MeitY) integrates broader enforcement via amendments to the Information Technology Act, 2000.
The Disclosure Regime
The Indian Procedural Law commences on the assumption that pleadings are professionally styled by the advocate who puts their signature upon them. Order VI Rule 14 CPC requires that pleadings bear the signatures of both the litigant and the pleader, thereby delineating a division of responsibility. Thereby, it is assumed that factual assertions are verified by the party and the legal formulation by the advocate. Regulation 43 fundamentally updates this assumption for the AI era. Instead of a blanket prohibition on AI-assisted drafting, the parties are now required to disclose the extent of such assistance, the verification undertaken before filing or whether any synthetic data or information was employed. Thus, the novel regulation recognised the unscrupulous result of AI usage, which may lead to the distortion of reliability and therefore affect procedural fairness.
The said regulation acts as an extension of ethical duties prescribed to advocates under the Bar Council of India Rules. Part VI, Chapter II requires advocates to act fairly, avoid misleading the court, maintain candour and employ only fair and lawful means. The submission of fabricated AI authorities would violate these duties. The regulation concomitantly complements existing ethical obligations under Section 35 of the Advocates Act 1961 (which lays down liability for misconduct), thereby transforming professional responsibility from an ex post disciplinary model into an ex ante compliance mechanism. Beyond user regulation, the draft also lays down obligations on AI vendors by mandating prior approval of judicial AI systems, disclosure of data governance practises and compliance with privacy safeguards under the Digital Personal Data Protection Act, 2023 and the Information Technology Act, 2000.
Thus, taken together, these provisions establish an inclusive framework ensuring accountability, wherein the significance of human oversight is elevated along with the regulated and verified usage of artificial intelligence.
Areas of Concern in the Draft Regulations
The draft regulation is a prudent intervention, but there exists an infirmity that needs to be addressed. It sets out broad principles, but often fails to detail the mechanisms that are required to achieve meaningful accountability. The foremost problem is a void in a comprehensive AI assurance framework. While the draft requires audits and disclosure documentation, there exists no minimum documentation requirement that any high-impact AI should keep. There is no requirement to keep version histories, audit logs, testing reports, records of updates or data sources as well. If those records are not available, you may be able to tell that AI was used, but you cannot tell if it worked correctly or if an error affected a judicial process.
The second problem is the draft’s presumption that courts possess sufficient technical knowledge to assess intricate AI models.
To preserve judicial independence, the plan heavily relies on internal court authority to audit AI systems. Contemporarily, many courts lack such expertise, thus making audits a procedural requirement without delivering the required effectiveness. Administrative uncertainty is another significant issue that the draft faces. There is no fixed timeline prescription through which an AI application fall outside the illustrative list provided in Regulation 19 could be provided.
While the proposal itself encourages courts to look for suitable alternatives under Regulation 16 and Regulation 17, a delay in approval might result in procedural bottlenecks. However, grievance mechanisms are provided, ideally. A person should not have to discover the role of AI after something goes wrong.
Affected parties must be informed when AI helps with court activities (such as translation, transcription, and case listing), given the opportunity to make corrections, and, if needed, granted access to human review.
While Regulation 43 requires lawyers to disclose the use of AI before the courts, there has been no issuance of any such direction or comprehensive rules by the Bar Council of India. More institutional coordination between the judiciary and Bar Council is imperative. The transparency provisions also require improvement. The Draft establishes reporting obligations and registries; however, it is unclear what data should be made public
Without jeopardising judicial security, a public record with basic information about the AI system’s purpose, the court in which it is being utilised, its risk category, and whether it has been audited would boost public trust.
Looking Ahead: Strengthening India’s Judicial AI Framework
India’s draft regulation represents a significant move, ensuring that technological advancements do not overshadow constitutional principles. The draft establishes that justice should not be assigned to machines while skillfully acknowledging the clear advantages of artificial intelligence in the legal sector. As artificial intelligence becomes more integrated into judicial processes, the challenge is to reconcile its use with the complexities of due process. This draft regulation lays a solid groundwork, but the critical task moving forward is to guarantee that efficiency does not compromise justice and that technology serves the rule of law rather than becoming an institution itself.
* The author is a fourth-year student pursuing a B.A. LL.B. (Hons.) at National Law University Jodhpur. The author may be contacted at anjali.gautam@nlujodhpur.ac.in
** The author is a fourth-year student pursuing a B.A. LL.B. (Hons.) at National Law University Jodhpur. The author may be contacted at 10336.uditjain.smjps@gmail.com
This blog reflects the personal views of the author and does not necessarily represent the views of The Policy Chronicle.