Synopsis
The Supreme Court has released a draft framework for AI in courts, aiming to balance innovation with judicial safeguards. The proposed regulations emphasise AI as an assistive tool, strictly prohibiting its use for decision-making, adjudication, or sentencing, while encouraging its adoption for administrative efficiency and access to justice.Listen to this article in summarized format
Lawyers from Cyril Amarchand Mangaldas, AZB & Partners, and Obhan Mason backed the framework’s core principle that AI should remain an assistive tool and never replace judges in decision-making.
While some foreign jurisdictions, particularly China, have experimented with AI-led decision-making in limited matters such as petty fines, those systems remain largely untested and unsuitable as a model for courts dealing with complex questions of rights and justice, lawyers said.
Replacing judges with AI tools would be problematic because of issues such as bias, hallucinations and the lack of transparency in AI systems, they warned.
The draft "Regulations for Use of Artificial Intelligence in Courts, 2026", released by the Supreme Court's AI Committee for public consultation, seeks to establish principles of human primacy, transparency, accountability, data protection and judicial independence in the deployment of AI across India's judicial system.
Stakeholders have been invited to submit comments by June 20.
The proposed regulations explicitly state that AI systems must remain "strictly subservient" to human judgment and judicial authority and can function only in an assistive capacity. The ultimate authority to determine matters of law, fact and justice would continue to rest exclusively with judicial officers.
The framework adopts a pro-innovation stance, creating a "presumption in favour of responsible AI adoption" and encouraging courts to actively deploy AI tools that improve access to justice, reduce delays and enhance administrative efficiency. However, it simultaneously prohibits the use of AI for dispute-outcome prediction and any replacement of human decision-making.
The draft regulations permit AI use in areas including legal research, precedent retrieval, citation verification, document summarisation, translation, transcription, case management, backlog monitoring and accessibility services, all subject to human oversight and verification.
At the same time, the regulations impose absolute prohibitions on AI-based adjudication and sentencing, risk-scoring systems, behavioural prediction of litigants or accused persons, opaque "black box" systems affecting rights or liberty, and AI-driven surveillance of judges, lawyers or litigants. These prohibitions are designated non-derogable and cannot be relaxed by any authority.
Legal experts welcomed the attempt to balance innovation with judicial safeguards.
Arya Tripathy, partner at law firm Cyril Amarchand Mangaldas, said the draft appropriately distinguishes between judicial decision-making and assistive functions.
“The draft guidelines maintain distinction between judicial decision making and assistive functions and are aimed at striking the delicate balance,” she said. While some foreign jurisdictions, particularly China, have experimented with AI-led decision-making in limited matters such as petty fines, those systems remain largely untested and unsuitable as a model for courts dealing with complex questions of rights and justice, she added.
Tripathy warned that replacing judges with AI tools would be problematic because of issues such as bias, hallucinations and the lack of transparency in AI systems. “Judicial decision making requires reasoned outcomes that have been delivered through due process, which raises fundamental questions on whether AI that suffers from blackbox can ever be compared with a human judge,” she said.
Tripathy said these provisions could help address some of the structural inefficiencies contributing to judicial delays. However, she cautioned that even AI-assisted legal research could subtly influence judicial outcomes if judges rely excessively on machine-generated recommendations.
“A bigger problem for AI in judiciary is confirmation bias,” she said, adding that AI systems trained on existing precedents may reinforce prevailing legal interpretations while overlooking novel circumstances that require human empathy and judgment.
Hardeep Sachdeva, senior partner at AZB & Partners, said the timing of the draft regulations was significant amid growing dependence on AI technologies.
"In today's time of increasing over dependence on AI, timings couldn't have been better for the SC's draft regulations striking a careful balance between judicial innovation and judicial independence," Sachdeva said.
He said that AI should remain an assistive tool for both lawyers and judges. "The framework rightly safeguards against any erosion of human analysis and judicial authority," he said, adding that even AI-powered legal research and summarisation tools require vigilance because subtle algorithmic biases could influence judicial reasoning.
Sachdeva also backed the draft's outright ban on AI-led adjudication, sentencing and risk scoring, calling them essential protections for due process rights and constitutional guarantees.
Essenese Obhan, managing partner of Obhan Mason, said the guidelines establish a clear policy favouring responsible adoption of AI while ensuring decision-making authority remains with judges.
"The guidelines articulate a clear policy in favour of the active adoption of AI within the judiciary, while confining AI to a strictly assistive role and reserving the decision-making power of the judicial officer," Obhan said.
He said the safeguard preventing AI from replacing judges was "soundly framed", particularly because the regulations place liability and verification responsibilities on the user rather than the technology itself. However, he cautioned that the effectiveness of the framework would depend on whether judicial officers genuinely verify AI-generated outputs rather than treating verification as a procedural formality.
The draft regulations make judicial officers accountable for decisions made with AI assistance and specify that AI-generated outputs must be treated as advisory. Courts are required to verify such outputs before using them, and accountability cannot be avoided by citing hallucinations or opaque AI systems.
Obhan said that AI-assisted legal research and summarisation would inevitably influence judicial outcomes because of potential biases embedded in training data. "Independent verification of outputs by the judicial officer therefore remains the practical safeguard against undue influence," he said.
The proposed framework also establishes a comprehensive institutional architecture, including a permanent apex body at the SC to oversee AI governance, dedicated AI committees in the SC and HCs, periodic audits, incident reporting systems, transparency disclosures and an AI content verification authority for generative AI outputs.
On global comparisons, Obhan said India's proposal stands out for its comprehensive nature. While the EU classifies judicial AI systems as "high-risk" under the EU AI Act, the UK relies largely on judicial guidance, Singapore has issued usage guidance for court users, and the US operates through a patchwork of court-specific directives.
"Against this backdrop, the Indian draft is a more mature and forward-looking step. It is the only one to be comprehensive, to be a binding regulatory code rather than be soft guidance or ad hoc orders, and the only one to lay down clear, absolute and non-derogable prohibitions while still embracing innovation," he said.
The consultation marks one of the most ambitious attempts globally to create a dedicated regulatory framework for AI in the judiciary, lawyers said.
