In September 2025, the National Company Law Tribunal admitted an insolvency petition against Essel Infraprojects Limited relying on legal precedents that, it turned out, did not exist. Some citations pointed to real Supreme Court judgments with fabricated paragraphs attached. Others pointed to cases that had never been decided at all.
When the matter reached the Supreme Court, a bench of Justices P.S. Narasimha and Alok Aradhe set aside both the NCLT and NCLAT orders, calling the reliance on fake, AI-generated precedents invisible, insidious, and catastrophic to judicial determination. The Court declared a zero-tolerance policy toward unverified AI-generated material in judicial proceedings and directed the Bar Council of India to examine the issue. That single case, more than any policy paper, explains why India's Supreme Court has now drafted its first comprehensive rulebook for artificial intelligence in courts.
On June 3, 2026, the Supreme Court's Artificial Intelligence Committee released the Draft Regulations for Use of Artificial Intelligence in Courts, 2026, for public consultation. The comment window, originally set to close on June 20, was extended to July 15, 2026, which has just passed as of this writing. The framework is not yet finalised or notified, but its shape is now clear, and it will affect every lawyer, court, and legal technology company operating in India.
"The lawyer who signs the pleading remains accountable for everything in it, whether a human or a machine produced the first draft."
Draft published by the AI Committee
Extended once from the original deadline
Supreme Court, High Courts, tribunals, districts
Headed by Justice P.S. Narasimha
This did not emerge from nowhere. The Supreme Court has been quietly building AI into its own registry functions for some time, from Su Sahayak, a website assistant for litigants, to SUPACE, a research tool used to help judges organise case material, alongside AI-assisted transcription of arguments and translation of judgments into regional languages. None of that experimentation had an overarching framework governing it until now. The draft regulations are as much about formalising what the Court had already started doing internally as they are about responding to the Essel Infraprojects episode.
The Incidents That Made This Regulation Inevitable
The Essel Infraprojects case was not an isolated embarrassment. It was the most visible entry in a pattern that has been building across Indian courts for several years, one that made some form of formal regulation almost unavoidable.
Essel Infraprojects: Fake Precedents at Tribunal Level
The NCLT and later the NCLAT relied on fabricated and non-existent judicial precedents to admit an insolvency petition. The Supreme Court set aside both orders and called for zero tolerance on unverified AI material.
Delhi High Court Refuses ChatGPT-Sourced Arguments
In a trademark dispute, the Court declined to rely on arguments generated through ChatGPT, observing that large language models were capable of fabricating authorities and factual material.
Bombay High Court Fines a Litigant Over Fake Citations
The Court discovered incorrect, AI-generated citations in a tax assessment matter and imposed a fine, treating the fabricated authorities as a serious lapse rather than a harmless error.
Punjab and Haryana High Court Bans Judicial AI Use
Facing mounting concerns, the High Court issued a circular barring all judicial officers from using AI tools, including ChatGPT, Gemini, and Copilot, for legal research or judgment writing.
Taken together, these incidents describe a judiciary that had already started using AI informally, without any shared rulebook for what counted as acceptable use, adequate verification, or professional accountability. The Draft Regulations are the Supreme Court's attempt to close that gap before it grows further.
What the Draft Regulations Actually Say
The Draft Regulations run to roughly thirty pages and were prepared by the Supreme Court's AI Committee, chaired by Justice P.S. Narasimha and including Justices Sanjeev Sachdeva, Raja Vijayraghavan V, Anoop Chitkara, and Suraj Govindaraj. They are built around five stated principles: human primacy, transparency, accountability, data protection, and judicial independence. Regulation 4 makes the first of these explicit, stating that human authority and judgment shall govern all functions, with AI restricted to an assistive role and matters of justice, fact, and law vesting completely in judicial authority.
Once notified, the regulations will apply to the Supreme Court, every High Court, and every district court, tribunal, and statutory adjudicatory body performing judicial or administrative functions in India. They will take effect on dates to be separately notified by the Supreme Court and by each High Court individually, so the actual rollout is likely to be staggered rather than instantaneous.
Human Primacy
AI may assist, but the authority to decide questions of law, fact, and justice rests exclusively with judicial officers, in every circumstance, without exception.
Transparency
AI use must be disclosed, and black-box or unexplainable systems are barred from any proceeding that could materially affect a party's legal rights.
Accountability
Responsibility for AI-assisted content stays with the human who filed or relied on it. AI cannot be blamed for an error a lawyer or court failed to verify.
Data Protection & Judicial Independence
AI use must comply with the Digital Personal Data Protection Act, 2023, and must never be structured in a way that compromises a judge's independent reasoning.
What AI Can and Cannot Do in an Indian Courtroom
Regulation 19 sets out permissible uses, subject to approval and supervision. Regulation 20 sets out prohibited uses, described in the draft as absolute and non-derogable, meaning no authority under the regulations can relax or waive them.
| What The Lawyer Handles | Without a Connected System | With LegalOS |
|---|---|---|
| Legal research and citation checks | Not addressed as a ban | Permitted, subject to verification |
| Drafting assistance and summarisation | Not addressed as a ban | Permitted, with mandatory disclosure |
| Translation and transcription | Not addressed as a ban | Explicitly permitted |
| Case management and scheduling | Not addressed as a ban | Explicitly permitted, including cause-list prep |
| Deciding a case or determining bail | Prohibited, absolute and non-derogable | Not permitted under any circumstance |
| Assessing witness credibility | Prohibited, absolute and non-derogable | Not permitted under any circumstance |
| Risk scoring of parties or accused persons | Prohibited, absolute and non-derogable | Not permitted under any circumstance |
| Surveillance of judges, lawyers, or litigants | Prohibited except if authorised by law | Requires explicit legal authorisation |
The pattern is consistent throughout the draft. Anything that speeds up the mechanics of legal work, research, drafting, translation, scheduling, is welcomed, provided it stays visible and verifiable. Anything that touches the actual exercise of judgment, deciding guilt, assessing credibility, predicting behaviour, is walled off entirely, with no carve-out large enough to erode the principle.
A New Layer of Institutions to Govern Judicial AI
The Draft Regulations do not just set rules, they create an entire governance apparatus to enforce them. At the top sits a permanent Apex Body at the Supreme Court, comprising two Supreme Court judges nominated by the Chief Justice, two High Court Chief Justices, two High Court judges, a representative of a reputable institution, a Joint Secretary-level officer from the Ministry of Electronics and Information Technology, and independent experts in finance, cybersecurity, data privacy, and AI.
The Apex Body
Sets minimum mandatory standards for AI use across all Indian courts, approves specific AI systems for deployment, and issues implementation guidelines that every court must follow.
Five Standing Committees
A Judicial Committee, Technical Committee, Committee on Infrastructure and Finance, Case and Data Management Committee, and Cyber Security Committee each advise the Apex Body within their domain.
AI Committees and Secretariats in Every High Court
Each High Court is required to set up its own AI Committee and a dedicated Secretariat to supervise local implementation and handle incident reports from within its jurisdiction.
CoRE-AI
The Centre of Research and Excellence on Artificial Intelligence provides ongoing technical and legal research support to the Apex Body as standards evolve.
The AI Content Verification Authority
A dedicated body charged with maintaining verification standards, tools, and protocols specifically for generative AI output used anywhere in the court system.
This is, in effect, a certification regime. Rather than each court or bench independently deciding whether a given AI tool is trustworthy, tools will need to pass through the Apex Body's approval process before they can be used in any official court function. For legal technology vendors, this approval pathway is likely to become the single most consequential regulatory gate in the Indian market.
How India's Approach Compares to Other Judiciaries
India is not the first judiciary to formalise rules for AI, but it is attempting something more detailed than most of what has come before it. Courts in France, Austria, and Greece have already relied on AI for adjudicatory and administrative functions in narrower, more targeted ways, generally focused on document translation, case triage, or anonymisation of judgments rather than a comprehensive governance framework spanning every level of the court system.
What sets the Indian draft apart is its institutional depth. Rather than a short policy circular, it proposes a standing Apex Body, five specialist committees, a dedicated research centre, and a content verification authority, all created specifically to keep pace with a technology that changes faster than most legal frameworks are built to handle. Whether that scale of institution-building is a strength or a source of bureaucratic drag will likely become clearer only once implementation actually begins.
Regulation 43: The Disclosure Requirement Every Lawyer Needs to Know
Buried in Chapter VIII of the draft is the provision most likely to reach every practising lawyer directly, whether or not their court has formally notified the regulations yet. Regulation 43(3) requires that any party or legal representative who uses AI in preparing pleadings or submissions must explicitly disclose the AI-assisted nature of that material.
The mechanism proposed is a filing provenance record: a structured disclosure identifying which AI tool was used, what sources it drew on, and what human review the output received before it was filed. The draft defines AI hallucination directly, as the generation of fictitious or inaccurate legal content, including fabricated precedents, statutory provisions, and quotations, and extends to lawyers the same standard of accountability the Essel Infraprojects ruling already applied to tribunals.
This is a significant departure from how most Indian lawyers currently use AI tools, which is largely informal and undisclosed. Once a version of Regulation 43 takes effect in a given court, undisclosed AI use in a filing stops being a private workflow choice and becomes a compliance question, one the Bar Council of India has already signalled it intends to scrutinise given its statutory authority over advocate conduct under the Advocates Act, 1961.
What This Means for Legal Tech Companies Operating in India
Apex Body Approval as a Market Gate
Once the approval pathway is live, tools that cannot demonstrate explainability, auditability, and verified sourcing will struggle to get anywhere near an actual courtroom workflow, regardless of how capable they otherwise are.
The End of the ChatGPT-in-Chambers Era
General-purpose AI assistants with no legal-domain grounding and no citation verification are precisely what triggered this regulation. Products built on that model face the most direct exposure.
Audit Trails Move From Nice-to-Have to Mandatory
Tools that automatically generate the disclosure and provenance records Regulation 43 will require become materially more attractive to firms trying to stay compliant without extra manual work.
Walled-Garden, Domain-Trained Systems Gain Ground
Platforms trained specifically on verified Indian legal sources, with every output traceable back to an actual statute or judgment, are structurally better positioned than tools that generate first and verify never.
How Lawyers and Firms Should Prepare, Starting Today
The regulations are not yet notified, and their final form may shift based on the feedback the Committee has just finished collecting. That is not a reason to wait. The direction of travel is unambiguous, and firms that adjust early will face a far smaller transition than those that wait for a mandate.
Audit Current AI Usage Across the Practice
Most firms have informal, undocumented AI use scattered across individual lawyers and associates. Know what tools are actually being used before a disclosure requirement forces the question.
Separate Verified Tools From Generic Ones
General-purpose chat assistants without legal-domain grounding are the highest-risk category under a regime built explicitly to catch fabricated citations.
Build the Habit of Disclosure Now
Start noting, internally, when and how AI assisted in a draft or research task, even before any court requires it. The habit is far easier to build gradually than to impose overnight.
Watch for Notification Dates Court by Court
Because rollout will be staggered by the Supreme Court and each High Court separately, the compliance clock will not start on the same day everywhere. Track your own jurisdiction specifically.
The Open Questions the Draft Doesn't Fully Answer
Can Supreme Court Rules Bind High Courts?
Chief Justice Surya Kant has flagged that High Courts hold broad constitutional powers under Articles 226 and 227, raising a real question of whether Supreme Court-issued AI rules can extend to High Courts and district courts without separate adoption.
Infrastructure Readiness Gap
Given well-documented capacity constraints across district courts, expecting appropriate AI infrastructure and oversight to emerge quickly at that level is a genuinely open question, not a settled assumption.
Data Security on Non-Indian Tools
Courts have historically leaned on foreign platforms like Zoom and YouTube for hearings and streaming, without clear data security rules, a gap the draft does not fully resolve for AI tools built on similar infrastructure.
Access for Unrepresented Litigants
The grievance redressal process under Regulations 52 and 53 assumes a level of legal literacy that a district court litigant without representation may simply not have.