Draft Regulations for Use of Artificial Intelligence (AI) in Courts, 2026

By 2026 the Supreme Court had two separate problems with AI. One was the courts' own tools: SUPACE for research assistance, SUVAS for translation, AI-assisted transcription and, from 2026, plans for AI-based listing and pendency tracking. The other was a run of cases in which lawyers, tribunals and trial judges cited judgments that did not exist. On 5 December 2025 a bench led by Chief Justice Surya Kant declined to issue directions on a petition seeking regulation of AI in the judiciary, saying that judges were being "extremely cautious" and that AI would never be allowed to override judicial decision-making, but invited suggestions on the administrative side.¹⁰¹¹ Within a week Bar & Bench reported that the Chief Justice had reconstituted the Supreme Court's AI Committee under Justice P.S. Narasimha.⁸

The first signal of a national rulebook came in the Court's hearings in the Gummadi Usha Rani matter. On 5 May 2026 the Bench told counsel that "the AI committee is looking into it" and was "almost at the verge of finalising the draft regulations", which would then be put out for consultation.⁵ The draft followed four weeks later.

The notice of 3 June 2026 states that a "preliminary draft" had been prepared "under the aegis of the Artificial Intelligence Committee" to govern AI in courts, "grounded in the principles of human primacy, transparency, accountability, data protection, and judicial independence", and invited comments from "all stakeholders and the general public" by 20 June 2026.¹²

Principles (Chapter II). The draft requires AI to be strictly subservient to human judgment (Reg. 4), consistent with the Constitution and natural justice and with the Bangalore Principles of Judicial Conduct (Reg. 5), non-discriminatory (Reg. 6), explainable (Reg. 7), auditable (Reg. 9), compliant with the Digital Personal Data Protection Act, 2023 (Reg. 10), limited to approved purposes (Reg. 11), proportionate with heightened safeguards for liberty-affecting uses (Reg. 12), and inclusive (Reg. 13). Reg. 8(3) treats AI output as advisory and requires reasonable verification, although an accountable officer may dispense with verification for reasons recorded in writing, and certified administrative tools may be deemed verified on a class basis.¹

Permitted and prohibited uses (Chapter III). Reg. 19 lists eleven illustrative permitted uses, subject to written approval and human oversight, from case management and transcription to anonymisation of judgments and backlog analytics. Reg. 20 lists nine prohibitions that are "absolute and non-derogable". They bar the use of personal data to train or refine AI systems without approval, outcomes reached through algorithmic decision-making alone, any AI system performing adjudication or sentencing without human-in-the-loop, risk scoring "for any purpose in Court processes, including the assessment of flight risk, prediction of recidivism, evaluation of bail eligibility, or determination of the credibility of parties or witnesses", opaque systems that may affect rights or liberty, profiling or prediction of conduct, surveillance of judges, advocates or litigants, undisclosed AI-generated evidence, and any use that compromises the confidentiality of deliberations.¹

Chapter IV proposes a permanent Apex Body at the Supreme Court, chaired by a Supreme Court judge and including two High Court Chief Justices, two High Court judges, an officer of Joint Secretary rank from MeitY, finance and cybersecurity experts, advocates with technology-law expertise and the head of AI at the National Judicial Academy, Bhopal. It would set minimum mandatory standards, approve AI systems, hear grievances, liaise with MeitY, NIC, the Data Protection Board and CERT-In, and publish an Annual Governance Report (Regs. 22-24). A Centre of Research and Excellence on AI would evaluate tools and publish research (Reg. 32). Every High Court would have an AI Committee of judges and an AI Secretariat headed by a District Judge (Regs. 33-34).¹²

Chapter V requires a Technical and Ethical Impact Assessment before any system is approved (Reg. 35), optional controlled-environment testing (Reg. 36), an AI Register (Reg. 37), audits at least yearly that must be done in-house, with no sharing of source code or datasets with outside auditors (Reg. 38), an incident database (Reg. 39), a one-year review of systems already in use (Reg. 41), fall-back protocols (Reg. 42), an AI Content Verification Authority for generative content (Reg. 44) and annual transparency reports (Reg. 45). Chapter VI forbids private vendors from participating without prior written approval and prescribes contract terms on data ownership, audit access, indemnity and on-premise or sovereign-cloud hosting for sensitive judicial data (Reg. 46). Chapter VII applies the DPDP Act and IT Act and restricts transfer of sensitive judicial data (Regs. 47-48).¹

The most contested provision is Reg. 43. Sub-regulation (3) says that where a party or lawyer uses an AI tool "in the preparation or submission of any document, pleading, or evidence, the AI-assisted character of such material shall be disclosed to the Court at the time of submission", by a declaration in a prescribed format. Sub-regulation (4) lets a court require disclosure of the system, the nature and extent of assistance and the verification steps. Sub-regulation (6) provides that if a filing is "fabricated, false, misleading, or inaccurate by reason of its AI-generated character, the person submitting the same shall bear full responsibility therefor and shall not be entitled to rely upon the character of the AI output as a defence".¹

The provision responds to the cases that prompted the draft. In Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. (2 July 2026) the Supreme Court declared that citing unverified AI-generated precedents is advocate misconduct and told the Bar Council of India to prescribe disciplinary norms, noting that regulations were being deliberated.⁶ Reg. 43(6) would write the no-defence principle into the courts' own rules.

The Supreme Court Advocates-on-Record Association (SCAORA) submitted comments reported on 19 July 2026. It called the disclosure requirement in Regs. 43(3) and 43(4) "unjustified" because advocates are already personally responsible for the accuracy of every filing under the Advocates Act, 1961 and the Supreme Court Rules, 2013, said routine disclosure would be impractical and could prejudice litigants or influence the judge, and proposed instead an amendment of the SLP certificate so that advocates certify personal verification of citations and authorities. It urged a phased, more precautionary rollout in place of "innovation over restraint", asked for audits of tools already in use, naming SUPACE, SUVAS, SuSahayak, AI-assisted transcription, e-SCR and AI-enabled e-filing, sought a seat on the Apex Body and clearer definitions of terms such as "Court Data" and "High-Risk AI Tools", and criticised the governance structure as overly bureaucratic.³

On 10 August 2026 Delhi High Court Justice Prathiba M. Singh was reported as saying that AI disclosure would make lawyers' lives difficult and could become "another set of templated affidavits", and as urging the Supreme Court to build an AI platform exclusively for the judiciary.⁴ The Kerala High Court, whose own policy for district courts is stricter on cloud chatbots, posted a notice on 10 June 2026 inviting comments on the draft.⁹ The draft's preference for responsible adoption contrasts with the High Court policies of Kerala, Gujarat and Punjab and Haryana, which on their terms restrict AI more tightly.

As of 15 October 2026 the regulations remain a draft. The comment window closed on 20 June 2026. On 2 July 2026 the Supreme Court said regulations were "being deliberated" and "will be notified after due process and in due course",⁶ and on 2 September 2026 a Bench of Justices Dipankar Datta and Sheel Nagu recorded that "the regulations are yet to be finalised" while setting aside a customs penalty order that relied on AI-hallucinated case law.⁷ The sources reviewed did not show a final text, a notification date or the constitution of the Apex Body. Under Reg. 1 the rules would take effect separately in each court by notification of its Chief Justice, so their reach would depend on adoption by individual High Courts.