Order in the Digital Court

On 3 June 2026, the Supreme Court invited comments and suggestions on the draft Regulations for Use of Artificial Intelligence (AI) in Courts, 2026 (Draft Regulations) prepared by the Artificial Intelligence Committee, headed by Justice P.S. Narasimha. 

The timing is not incidental. The Draft Regulations are in continuance of the Court’s push to adopt AI in its registry functions, with the acknowledgement that justice must keep pace with technology without ceding ground to it. The Supreme Court of India now joins the list of Courts across the world such as France, Austria and Greece that have relied on AI for adjudicatory and administrative functions. 

The Court has already administered AI in its functions for argument transcripts, translation of judgements, providing website assistance using Su Sahayak and SUPACE without an overarching framework to govern their use. 

Conversations around ethical AI-use floated when a trial court was found to have relied on a fabricated AI-generated precedent earlier this year. In July, the Court identified six “hallucinated” judgements in an Order passed by the National Company Law Tribunal. Against this backdrop, the draft regulations aim to place principles, permissions and prohibitions on AI use in courts. 

What AI Regulations introduce?

Two days after the Draft Regulations were made public, Chief Justice Surya Kant, in London, stated that technology is neither “inherently benevolent nor inherently harmful”. He stressed that technological power must remain “accountable to constitutional values, democratic legitimacy and human dignity”. The Draft Regulations have reiterated that AI use must be ethical and “in a responsible manner”. Chapter II, titled “General Principles to Govern Adoption, Deployment and Use of AI Systems”, states that AI should be strictly subservient to human judgement “at all times”. Further, AI should be administered in an assistive capacity and should not replace autonomous human decision making. 

The same chapter ties AI use to principles of natural justice, fairness, and equality before law. It bars the deployment of AI systems perpetuating, amplifying and introducing discrimination on the grounds of race, religion, caste, sex, gender, disability, language, economic status, besides any other grounds prohibited under the Constitution. The chapter frames individual and institutional accountability where judicial officers are supposed to bear all responsibility for AI-based decision making. Every AI-generated output or information must be verified before use, with the exception that an officer may, for reasons being recorded in writing, dispense with the requirement of verification. The regulations put in place for continuous monitoring and timely audit of deployed AI systems.

The objective of the Regulations

Chapter III details out the permissive use of AI, which though not exhaustive, includes areas which the Supreme Court has not charted before. Apart from transcriptions and translations, the regulations allow AI-assisted case management, cause list preparation, hearing scheduling and docket prioritisation. In his London address, the Chief stated that “such technologies can help reduce delays, improve efficiency, expand access to legal information, and enable judges and court administrators to focus their attention on the more nuanced and inherently human aspects of adjudication”.

The Draft Regulations appear to streamline much of the procedural nitty-gritties by deploying it in case filing, defect scrutiny and record management. The Appropriate Authority has discretion to use AI in other matters with prior approval in writing. Taking cue from countries like Spain, Austria, France and Greece, the regulation permitted the use of AI for anonymisation of judgements and court records. AI tools will be deployed to redact personal information of parties, dates, addresses, etc. Regulation 20(1) lists out the prohibited use of AI and mandates, under 20(2), that the restrictions are absolute. 

AI use by the litigant

In February 2026, a Bench of CJI Surya Kant, Justices B.V. Nagarathna and Joymalya Bagchi had expressed its concern over lawyers and litigants using AI to draft petitions, adding that it is “absolutely uncalled for”. Moreover, the Bench stated that other sitting judges of the Supreme Court had noticed this with non-existent precedents being cited in the petitions. The Bombay High Court had fined a petitioner after discovering incorrect citations that were AI generated.

The Draft Regulations give recognition to this problem. It defines “AI hallucination” as the generation of fictitious or inaccurate legal content, including fabricated judicial precedents, statutory provisions and quotations. It extends similar responsibility and accountability as imposed on judicial officers to the lawyers. Regulation 43 requires a declaration to accompany filings where AI has been used. Anyone using AI has to exercise reasonable diligence in verifying the accuracy of the material generated. Whatever the technology may produce, the responsibility of relying on it is entirely human. 

The requirement brings the issue directly to the filing counter. Other jurisdictions such as England and Wales, Australia and parts of Canada place the burden on lawyers to verify AI-generated material but generally stops short of making disclosure a procedural requirement. 

The Regulations attempt to prevent that problem before documents enter the judicial record. Even so, the provision does not distinguish between different categories of software. It remains unclear whether the declaration requirement is confined to generative AI used for drafting and legal research or whether it extends to AI-assisted translation software, transcription platforms or increasingly common features such as predictive writing and grammar suggestions integrated into word-processing applications.

The practical implications may be felt most acutely in urgent matters. Applications seeking interim relief, mentioning matters or habeas corpus petitions are often prepared and filed within a matter of hours. Whether the additional declaration becomes a routine procedural formality or another point at which registry objections are raised will only become clear once the Regulations are put into operation.

Further, the Draft Regulations claim that “reasonable care” should be taken in verifying AI-generated material but offer little guidance on what that standard entails. The Regulations do not presently answer those questions, leaving them to future practice.

Can AI decide what gets heard first?

The Regulations reject the possibility that machines will decide cases. Administrative assistance does not determine the outcome of litigation. The Draft Regulations claim that AI will be administered in the listing of cases and “judicial resource alllocation”. This raises a question that the Draft does not directly address: how does AI-assisted listing sit with the Chief Justice’s role as the Master of the Roster?

Advocate-on-Record Eklavya Dwivedi believes the answer lies in understanding the source of that power. According to him, the Constitution does not expressly designate the Chief Justice as the Master of the Roster. Rather, the authority flows from the Supreme Court Rules and long-standing constitutional convention, a position recognised by the Supreme Court in Shanti Bhushan v Union of India (2018). In his view, carefully designed AI systems could improve transparency in a process that has often attracted criticism for concentrating administrative discretion in a single office. “This unilateral and unchecked power impedes the checks and balances principle the Constitution seeks to establish, as also the lack of transparency in allocation of matters to particular Benches,” he says. “In this regard, the use of AI can alleviate these concerns,” he adds.

That optimism is accompanied by an equally important caveat. AI can only assess the material placed before it, and pleadings do not always present a complete or neutral account of the dispute. “Scrutiny through AI tools would be limited to what is stated in pleadings, which experience shows is not always a fair representation of the matter. Lawyers often present a narrative which aligns with their clients’ interests,” Dwivedi says. If AI is eventually used to assist with listing, he suggests that courts may have to discourage inaccurate pleadings through the imposition of costs so that parties do not attempt to manipulate automated scrutiny.

His observation points to a broader limitation of AI in judicial administration—algorithms cannot independently verify whether a petition truly raises questions of personal liberty, constitutional significance or commercial urgency. Dwivedi adds, “The application of technological and AI tools ensures, at the very least, uniform scrutiny based on the codes that underpin the software. Judicial outcomes would largely depend on independent application of mind by the respective judge, and the AI interface should have no role to play in this regard.”

Asked what he considered the greatest danger in introducing AI into case listing, Dwivedi’s answer was restrained: “None that I foresee.” 

Who watches the algorithm?

Regulations 23 to 39 establish a layered system of oversight beginning with the Supreme Court Apex Body, which is tasked with framing policy, approving AI deployments and prescribing standards for courts across the country. Each High Court is required to constitute its own AI Committee, while day-to-day implementation is entrusted to an AI Secretariat headed by an officer not below the rank of District Judge. 

The Draft also proposes a Centre of Research and Excellence on Artificial Intelligence (CoRE-AI) to test and evaluate AI systems before they are introduced into judicial processes. The Regulations also require every approved AI system to be documented through an AI Register maintained by the relevant authority. The Register records the purpose for which the system has been deployed, version updates and other technical details. While the Regulation does not expressly state whether the Register will be publicly accessible, maintaining such records could become important where litigants seek to question the functioning of a particular AI system or challenge the manner in which it has been deployed.

Commercial AI models are updated frequently, often without public notice. A system approved after one round of testing may undergo several iterations within months. The Regulations therefore place considerable responsibility on the Apex Body, AI Committees and the Secretariat to ensure that approvals do not become a one-time exercise but continue throughout the lifecycle of the technology.

Regulation 52 creates a specific grievance mechanism where harm is caused by a prohibited use of AI. An affected party or legal representative may move the court in which the AI system was used and seek appropriate relief after an opportunity of hearing. 

Keeping court data inside the Court

Protection of personal information through data minimisation recurs across the Draft. The exception to the principle of data minimisation is that the AI system is allowed to collect and retain such personal data which is strictly necessary for the specific purpose for which it is deployed. Regulation 10 assures that the right to privacy will be ensured in all AI related operations, further the processing of any personal data will be in accordance with the Digital Personal Data Protection Act, 2023. 

The Draft Regulations introduce three familiar principles borrowed from data protection law: purpose limitation, data minimisation and privacy by design. AI systems may process minimum data necessary to perform that task and comply with the requirements of the DPDP (Regulation 47). However, the restrictions do not end there. Regulation 14 prohibits the deployment of AI systems trained on unlawfully collected or demonstrably biased datasets. AI systems are also prohibited from being used for risk scoring and predictive profiling of individuals. If the reliability of a particular AI system is challenged in future litigation, questions may arise about the provenance of the datasets on which it was trained and whether they satisfy the standards prescribed under the Regulations.

Regulation 48 adds a second layer of protection specifically for Sensitive Judicial Data, a category that includes personally identifiable information relating to parties, witnesses and legal representatives, as well as information generated during court proceedings whose unauthorised disclosure may cause harm. The Regulation requires periodic security audits, anonymisation measures and restrictions on the movement of sensitive judicial data outside authorised environments. The underlying concern is evident: court records cannot become training material for commercial AI systems simply because a private vendor develops software for the judiciary.

The Draft therefore contemplates contractual safeguards that would require technology providers to use judicial data only for approved purposes. If implemented in the form presently proposed, these provisions could require vendors to segregate court data from commercial model development and ensure that software developed with public judicial infrastructure remains subject to judicial control.

The Lines AI Cannot Cross

That Indian framework is on a different footing from the EU AI Act, which classifies many judicial AI applications as “high-risk” and regulates their use through enhanced compliance obligations rather than prohibiting them altogether. 

The prohibition on risk scoring and behavioural prediction reflects concerns that have surfaced in criminal justice systems elsewhere. Predictive assessment tools such as COMPAS, used in parts of the United States to estimate the likelihood of reoffending, have generated sustained debate over transparency, algorithmic bias and the inability of defendants to meaningfully scrutinise proprietary scoring methodologies. The Draft Regulations avoid that debate by excluding comparable applications from judicial decision-making altogether.

The same philosophy appears in the requirement that AI systems be explainable. Regulation 7 requires AI systems to be capable of generating an account of the reasoning or process by which an output was reached, in terms that can be understood by judges, court staff and litigants. While the Regulation does not create a standalone right for parties to demand an explanation of every instance in which AI has been used, it signals an unwillingness to permit “black box” technologies to influence judicial processes without meaningful human understanding.

More than a technology policy

The Draft Regulations arrive at a moment when courts across the world are experimenting with AI, but few have attempted to regulate its use in the level of detail proposed by the Indian Supreme Court. The document is neither an endorsement of AI nor an expression of distrust. Instead, it treats AI as a tool that may improve the administration of justice, provided its use remains transparent, accountable and subject to continuous human supervision.

The draft Regulations were open for public consultation until 20 June. The comments they receive will determine whether the final framework remains as proposed or changes to address concerns raised by judges, lawyers, technology companies and civil society. Much of that discussion is likely to center on how these rules will work in day-to-day court functioning, rather than whether AI has a place in the justice system.