A founder's note by Saakar Yadav, Founder and Managing Director, Lexlegis.ai
On 3 June 2026, the Artificial Intelligence Committee of the Supreme Court of India published a preliminary draft of the Regulations for Use of Artificial Intelligence in Courts, 2026, and invited comments from the public and from stakeholders by 20 June 2026. At Lexlegis.ai, we read it as one of the most important documents to emerge for our field in recent memory. We welcome it without reservation.
For the better part of three decades, the people behind Lexlegis.ai have worked on a single idea: that technology, used carefully, can help justice move faster without compromising the things that make justice trustworthy. That journey runs from an early search engine for tax decisions in the late 1990s, to building the National Judicial Reference System for the Government of India, to the legal language model we operate today. Across all of it, our conviction has stayed the same. The goal is not to impress with what a machine can generate. The goal is to expedite justice in a way that practitioners, judges, and citizens can rely upon. The Court's draft speaks directly to that conviction.
What stands out most is the balance the draft strikes. It does not treat artificial intelligence as a threat to be contained, nor as a novelty to be adopted uncritically. It treats it as a tool to be governed. The framework is built on five clear principles: human primacy, transparency, accountability, data protection, and judicial independence. Around those principles, the draft adds something many policy documents lack: a working presumption in favour of responsible adoption, and an explicit preference for innovation over restraint where the safeguards are met.
That combination is rare and, in our view, exactly right. It tells builders that the door is open, and it tells them precisely what they must carry through it.
Several provisions read as if they were written for the way reliable legal AI ought to be built in the first place.
The draft keeps the human in command. AI is to remain strictly assistive and advisory, with the final decision, and the accountability for it, resting with the judicial officer. We have always designed to this standard. Our systems exist to help a lawyer or a judge work with greater clarity and speed, never to substitute for legal judgment.
The draft demands transparency and explainability, and it restricts opaque systems in matters that affect liberty or rights. This is the heart of how we build. Our platform is grounded in an authoritative legal corpus, every answer is linked to its sources, and the reasoning is presented so that a user can see how a conclusion was reached and test it. An answer that cannot be traced to legal authority has limited value in a courtroom, and the draft recognises this plainly.
The draft confronts hallucination by insisting on verification and on disclosure of AI-generated and GenAI content. Anyone serious about legal AI knows that a confident, fluent, and entirely fabricated citation is the central risk of this technology. We addressed it by constraining the model to a verified corpus and reasoning over retrieved legal texts rather than relying on statistical memory, and by adding a layer of citation checking and cross validation above every output. The draft turns that engineering discipline into a shared expectation for the whole ecosystem.
The draft is careful about data. Data minimisation, anonymisation, sovereign or on-premise handling of sensitive judicial data, and strict limits on the use of court data are all part of the text. These are the right protections for information that belongs, ultimately, to the public and to the parties before the court.
For responsible companies, regulation of this kind is not a burden. It is a foundation. Clear rules let serious builders compete on the things that matter, namely accuracy, traceability, security, and genuine usefulness, rather than on marketing claims that cannot withstand scrutiny. The prohibitions in the draft, on algorithmic decision-making in place of a judge, on risk scoring, on profiling and surveillance, draw bright lines that protect citizens and, in doing so, protect the credibility of the entire field.
We also welcome the institutional design. An Apex Body, AI Committees, impact assessments, audits, registers, and incident reporting create a structure where technology can be tested, monitored, and improved over time, in controlled environments before it touches a live proceeding. That is how trust is earned, and trust is the only currency on which legal technology will endure.
With this draft, India places itself among the first major jurisdictions in the world to publish a structured, principle-led framework for the use of artificial intelligence in its courts. That is a significant statement. It signals that the country intends to lead on responsible AI in the justice system rather than wait for problems to force a response, and it does so from a position of strength, given the scale and diversity of India's judiciary.
The approach is also one that other emerging economies can study and adapt. A framework that pairs firm safeguards with an explicit openness to innovation offers a credible alternative to the two extremes of unchecked adoption on one side and reflexive prohibition on the other. For Indian legal technology, it sets a clear and demanding standard to build to, and a standard built at home can be carried abroad. We have seen, through our own work on national judicial infrastructure, that India is fully capable of producing public technology of global quality. This draft creates the conditions for more of it.
The draft does not speak only to courts and vendors. It speaks to practitioners. Where AI assists in preparing a document, pleading, or piece of evidence, the draft contemplates that the AI-assisted character of that material be disclosed to the court, and it places responsibility for accuracy squarely on the person who files it. The character of an AI output will not serve as a defence for a fabricated or misleading submission.
For law firms, in-house teams, and individual advocates, the practical lesson is straightforward. The choice of tool now carries professional consequences. Tools that are grounded in authoritative sources, that link every assertion to a verifiable citation, that show their reasoning, and that keep client and case data secure are no longer merely preferable. They are the means by which a practitioner discharges the duties this framework describes. Adopted well, AI of this kind lets a legal team work faster and with more confidence while reducing, rather than creating, the risk of error. That is precisely the outcome we build for.
Everything we have built at Lexlegis.ai serves one base objective, namely to accelerate justice. Faster does not mean looser. It means removing the delay that comes from manually searching vast bodies of law, reconciling conflicting authority, and reworking drafts, so that the time saved can be returned to judgment, which is the part that must remain human.
Our foundation rests on a simple structure that we describe as Ask, Interact, and Draft. With Ask, a user puts a legal question and receives an answer grounded in an authoritative corpus, with every point linked to its source. With Interact, a user works directly with documents, judgments, contracts, and case files, questioning them and drawing out what matters. With Draft, the user moves from research to a working document. Each of these maps cleanly onto the permissible uses the draft contemplates, including legal research, precedent retrieval, citation verification, summarisation, translation, and document preparation, and each is designed so that the human remains the one who decides.
MIRA, our Machine Intelligence and Reasoning Assistant, builds on this foundation. Rather than a single fixed tool, MIRA is configurable, so that an organisation can assemble only the capabilities a given matter requires, from research and drafting to administrative and compliance work. Above its outputs sits a reasoning and verification layer that performs citation checking and cross validation, so that what is produced can be traced and tested rather than taken on faith. This is our answer to the central risk of legal AI, which is fluent and confident text that is not supported by authority.
Read against the draft regulations, the alignment is direct. The framework asks for grounding, explainability, verification, disclosure, and a human who remains accountable. These are not features we are adding in response to a rule. They are the principles on which MIRA and the Ask, Interact, and Draft foundation were designed. A tool built this way does not merely comply with the draft. It advances the very purpose the draft exists to serve, which is a justice system that moves faster while remaining worthy of trust.
Lexlegis.ai intends to participate constructively in the consultation, and to align our products with both the letter and the spirit of these regulations as they take shape. We will continue to build systems that are grounded, explainable, and verifiable, that keep human judgment at the centre, and that treat the confidentiality of judicial data as a first principle rather than an afterthought.
The legal profession has always understood that justice depends on confidence, and confidence depends on authority that can be checked. The same is true of the technology that now assists it. The Supreme Court's draft holds AI to that standard. We are glad it has arrived, and we look forward to helping make it work.
Saakar Yadav is the Founder and Managing Director of Lexlegis.ai. Comments on the draft Regulations for Use of Artificial Intelligence in Courts, 2026 may be sent to the Member Secretary, AI Committee, Supreme Court of India, by 20 June 2026.