Something significant happened in Indian legal history on June 3, 2026, and most practicing advocates were too busy with their morning cause lists to notice. The Supreme Court of India released draft "Regulations for Use of Artificial Intelligence (AI) in Courts, 2026", a framework that covers every court, tribunal, and adjudicatory body in the country. Public comments closed June 20. The final version hasn't arrived yet. But the direction is unmistakable, and advocates who think this is a policy document for academics are making a serious professional miscalculation.

It is a liability framework dressed in the language of governance. And it will govern your practice far sooner than you expect.

The Disclosure Obligation: What It Actually Means

Strip away the committee language and Draft Regulation 43(3) says one thing plainly: if you used AI to prepare any pleading, document, submission, or piece of evidence, you must tell the court. Not in the footnotes. At the time of filing. You must also be ready to identify which AI system you used, how extensively you used it, and what you did to verify its accuracy.

The draft goes further than disclosure. If anything, AI-generated in your filing turns out to be false, fabricated, misleading, or inaccurate, you bear full professional responsibility. The fact that a machine produced the error is explicitly not a defence.

Read that again. You cannot outsource accountability to an algorithm. Not in India's courts. Not under these regulations.

For most advocates who have been quietly using ChatGPT to generate first drafts of writ petitions or running their facts through a generative tool to clean up the narrative, this is the end of the shadows. The disclosure regime makes AI-assisted drafting visible. The visibility changes everything about how you need to work.

Why This Urgency? Because It Has Already Gone Wrong.

In February 2026, just months before these draft regulations were released, the Karnataka High Court ordered a probe against a trial court judge for citing Supreme Court judgments that did not exist. The judgments were almost certainly AI-generated. The court had no idea. Legal Eagle reported the SC rap as recently as March 2026.

This was not an isolated incident imported from American headlines. It happened here. In our courts. Against one of our judges.

India's courts have also watched what has unfolded in the United States, United Kingdom, Canada, and Australia jurisdictions where lawyers have already faced sanctions, fines, and professional disciplinary action for submitting AI-hallucinated citations without verification. A Stanford HAI study found that legal AI tools specifically built and marketed for legal work hallucinate in more than one out of every six benchmarking queries. One in six. That is a recurring professional hazard.

Former Chief Justice B.R. Gavai publicly warned the profession about the risks of tools like ChatGPT generating fake citations. The Supreme Court is not issuing these regulations in a vacuum. It is issuing them because the problem has already arrived.

The Permitted Uses Are Genuinely Broad

The draft is not the work of technophobes. The permitted use cases are expansive: legal research, citation verification, drafting assistance, translation, transcription, case management, hearing scheduling, docket prioritization, record management, and litigant-facing information services. Draft Regulation 16 goes further still, it creates a presumption in favor of responsible AI adoption and directs courts to actively deploy AI tools that improve access to justice and reduce delays.

In other words: the Supreme Court wants AI in courts. It is not trying to stop the technology. It is trying to stop the recklessness.

AI will become more deeply embedded in legal practice as the courts deploy their own tools. SUPACE is already live for judicial research. SUVAS has translated over 36,271 Supreme Court judgments into Hindi. Adalat AI is recording proceedings in subordinate courts across Kerala with a mandatory High Court deployment order since November 2025, and it has cut case timelines by 30–50% in measured deployments. The direction is irreversible.

The Hard Red Lines You Cannot Cross

AI cannot be used to predict future conduct, which means bail applications cannot be assessed with AI risk scores. AI cannot evaluate witness credibility. Courts cannot use opaque "black-box" systems in any matter affecting rights or personal liberty. And no judicial outcome can be reached through algorithmic decision-making alone.

For advocates in criminal practice, this is a shield you can now invoke. If opposing counsel is using an AI-derived risk profile to argue against your client's bail, the regulations give you a framework to challenge it.

What the Governance Structure Signals

The draft proposes a permanent Apex AI Body at the Supreme Court level, AI Committees in every High Court, and a Centre of Research and Excellence on Artificial Intelligence (CoRE-AI) to evaluate tools and inform policy. Every AI system deployed in courts faces annual audits. An AI Incident Database will track errors and breaches. Private vendors can participate only with prior approval, and crucially, AI built using court data cannot become the exclusive IP of any private company.

What You Should Do Before the Final Rules Land

Start a verification habit now. Every AI-generated output you rely on — every case citation, every statutory reference, every doctrinal claim — must be confirmed against a primary source before it goes into a filing. Understand what you are using: the difference between Manupatra's native legal AI and a generic chatbot with a law prompt is the difference between a sourced citation and a fabrication risk. Treat disclosure as an opportunity, not a penalty.

The Supreme Court has acted with more foresight on this question than most regulators globally. It is not banning AI from the profession. It is demanding accountability from the humans who deploy it. The advocates who internalise it earliest will be the most dangerous ones in court.