From SUVAS to SUPACE: The Judiciary Adopted AI Before the Bar Did

The most consequential adoption of artificial intelligence in Indian law did not happen at a large firm or a legal-tech startup. It happened at the Supreme Court of India, and it began earlier than most of the profession realises.

On 26 November 2019 — Constitution Day — the then Chief Justice, S.A. Bobde, launched the Supreme Court Vidhik Anuvaad Software, SUVAS. It was not a general translation tool with a legal label. SUVAS is a domain-specific neural machine translation system, trained on judicial texts so that terms of art, procedural language and statutory phrasing survive the journey between languages. The problem it addresses is constitutional in origin: Article 348(1)(a) requires proceedings of the Supreme Court and the High Courts to be in English, in a country where the overwhelming majority of litigants do not read English. The English text remains authoritative; the translations are informational. But informational, at scale, changes access to justice.

And the scale arrived quickly. By 2024, SUVAS had been used to translate more than 36,000 Supreme Court judgments into Hindi and over 17,000 into other regional languages, published on the e-SCR portal; the Court now reports coverage across sixteen regional languages, with more planned. High Courts — Allahabad, Delhi, Kerala among them — have adopted the platform to publish their own judgments in regional languages. For a practising advocate, the practical upshot is simple and slightly startling: there is a fair chance the Supreme Court can already hand your client a judgment in your client's own language.

SUPACE followed in April 2021: the Supreme Court Portal for Assistance in Court Efficiency, a research tool that reads the factual matrix of a case and organises relevant precedent for the judge. CJI Bobde described it at launch as a "perfect blend of human intelligence and machine learning". Its design choice is the interesting part. SUPACE does not predict outcomes and does not recommend results; it assembles and organises material so the human can reason over it. The Government has been candid that the system remains in an experimental stage pending serious computing infrastructure — but the architecture is the message: in the judiciary's own house, AI is an assistant with clear limits, never a decision-maker.

The programme has since widened into a stack. TERES transcribes court proceedings — Constitution Bench hearings among them — into searchable text. Su-Sahayak answers litigants' questions about procedure, filing and case status through a conversational interface. A generative research assistant, LegRAA, is being developed for internal legal research. Behind all of it sits Phase III of the e-Courts project (2023--2027), with a total outlay of ₹7,210 crore, of which ₹53.57 crore is specifically earmarked for AI and blockchain integration across the High Courts. This is not a pilot. It is infrastructure.

Just as instructive as the building is the fencing. In July 2025 the Kerala High Court issued what is widely regarded as the first formal AI policy for the district judiciary in India, barring judicial officers from using AI tools to arrive at findings or to draft orders. In November 2025 the Supreme Court's Centre for Research and Planning published a White Paper on AI and the judiciary that named the fabrication of case law as a primary risk and mandated independent verification of anything an AI tool produces. And in June 2026 the Court released draft regulations on the use of AI in courts for public consultation — draft rules that would keep AI out of adjudication altogether and require disclosure when it is used. A governance structure has existed since 2019 in the form of the Court's AI Committee, first chaired by Justice L. Nageswara Rao.

What should a working advocate take from all this? Four things.

First, the legitimacy question is settled. Whether AI belongs in Indian legal work is no longer a debate the Bar needs to have; the country's highest court answered it by building its own. The live questions are narrower and more useful: which tasks, which tools, and under what discipline.

Second, the judiciary's operating model — assistive, human-accountable, verification-first — is the template for chambers as well. Notice what the courts automated first: translation, transcription, research organisation, procedural information. Not judgment. Not drafting orders. The same division of labour maps precisely onto practice: let the machine find, summarise, translate and transcribe; keep strategy, advice and final drafting in human hands.

Third, language expectations are shifting under the profession's feet. When the Supreme Court publishes its output in sixteen languages, a client will reasonably begin to expect the same courtesy from counsel — status updates, key orders and advice in a language the client actually reads. Advocates equipped to do this at the press of a button will feel the difference in their client relationships within months, particularly in mofussil and district practice.

Fourth, disclosure norms are coming. The draft regulations point clearly toward a regime in which the use of AI in filings must be verifiable and, in some form, declarable. The practical preparation is not to avoid AI — that window has closed — but to use tools whose every output you can trace to a real, openable authority and stand behind under your own signature. Tools that cannot show their sources will become professionally unusable the day the first disclosure requirement lands.

There is a comfortable irony in where this leaves the profession. The institution most often described as slow-moving has, on this subject, moved first and moved carefully — adopting the technology for what it does well, walling it off from what it must never do, and writing the rulebook in public. The Bar could do considerably worse than to follow its lead.

---

— LawWorld is built on the same principles the judiciary chose for itself: AI that retrieves from a real database of judgments, Acts and commentary, shows its sources, and leaves every decision with the advocate. Translation between English and regional languages, with legal formatting intact, is included in every plan.