India's Supreme Court moves to bench AI in its own courtrooms
Proposed rules would forbid AI from drafting orders or substituting for oral arguments, while permitting limited use for legal research and translation. Lawyers say the line between assistance and abdication is thinner than the bench admits.
On 13 July 2026, the Supreme Court of India published proposed regulations that would for the first time define, line by line, what artificial intelligence may and may not do inside the country's highest courts. The draft, summarised by The Indian Express, draws a sharp boundary: AI tools can be used for legal research, translation, and summarising voluminous records, but they cannot draft orders, simulate judicial reasoning, or stand in for oral argument. Any output that purports to predict a ruling is explicitly barred.
The framing matters because Indian courts are not debating whether AI will arrive in the courtroom. They are negotiating how visible the substitution will be. A bench that uses a model to triage 40,000 pending matters has made a different choice than one that uses the same model to draft the operative paragraph of a judgment, and the new rules are an attempt to lock in the first posture before the second becomes habit.
What the draft actually permits
The regulations, as reported by The Indian Express, sort AI use into three buckets. Permitted uses include searching case law, converting filings across India's many scheduled languages, checking citations against the court's own database, and producing first drafts of routine procedural orders, provided a human judge reviews and signs off. Conditionally permitted uses cover fact extraction from long documentary records and the preparation of case-brief summaries by court-appointed research staff. Barred outright: any system that generates the reasoning or operative portion of an order, any tool presented to a party as an arbiter, and any use that would replace oral arguments in person or by video.
The court also proposes a disclosure regime. Filings that rely materially on AI-generated text would have to carry a declaration identifying the tool and the manner of use. Failure to disclose, the draft warns, could attract contempt-style sanctions. The disclosure provision is the section most likely to do real work; the rest is largely a recitation of what most judges already believe they would not do.
The pressure the rules are responding to
Indian trial and appellate courts now handle caseloads that no Anglophone peer jurisdiction matches. The backlog is the structural reason AI is being invited in at all. Senior advocates and registry staff have told the bar repeatedly that the practical question is not whether machine assistance will scale, but who sets the floor under it. The bench that drafts these rules is, in effect, trying to set that floor before commercial vendors do. Several large Indian law firms have already internalised document-review pipelines built on commercial large language models; smaller chambers have followed. A rule book written by the apex court changes the incentives for both.
The other pressure is more political. India's draft framework for AI, the IndiaAI mission, treats compute capacity, Indic-language models, and sectoral deployment as one project. A judicial module, written by judges rather than the Ministry of Electronics and Information Technology, sends a signal to that project that the courts intend to be rule-makers rather than rule-takers in their own domain. The signal is the point, perhaps more than the substance of any one clause.
What the critics on the bar will say
Two lines of objection are predictable. The first, from senior counsel, is that disclosure requirements will chill legitimate use. Lawyers who lean on AI for translation and citation checking do not, on this account, want to brand every brief as machine-assisted when the contribution is mundane. A blanket declaration rule risks creating a stigma where none is warranted. The second, from court staff associations, is that the ban on AI-drafted reasoning will be unenforceable without a forensic audit trail of how each judgment was produced. If the line is internal to a judge's chambers, there is no practical way to police it short of surveillance the judiciary would not tolerate.
There is also a quieter critique, more sympathetic to the draft's intent. The court, on this read, is regulating the visible surface of AI use while leaving the deeper question untouched: whether a generation of judges trained on machine-generated research will develop the instincts to second-guess it. That is not a rule the bench can write. It is a question the legal academy will spend the next decade answering.
What to watch before the rules take effect
The draft sits open for comment before the full court takes it up. Three dates will tell whether the regulations harden or soften. First, the comment period itself and which sections attract the most objections from the bar councils and the attorney general's office. Second, the framing of any subsequent circular from the registry on AI-assisted translation in district courts, where the caseload burden is heaviest and the bench supervision thinnest. Third, the first reported instance in which a high court or tribunal cites these rules in an order, either to admit or to exclude a piece of AI-generated evidence. The line between those two outcomes will, in practice, become the line of the law.
The deeper uncertainty is whether the regulations will age well. Commercial AI products move faster than judicial consultation cycles, and the categories the draft carves out today (research, translation, summarisation) will look different once models that draft reasoning are rebranded as "advanced summarisation" by their vendors. The court is regulating a moving target with a fixed instrument, and the contest over what that instrument actually covers is the contest the next five years will be about.
This piece treats the Indian judiciary as a rule-maker with agency in its own digital transformation, not as a passive recipient of imported technology. The Western wire line on AI governance tends to centre Brussels and Washington; the Indian draft, drafted by judges rather than ministers, deserves to be read on its own terms.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://en.wikipedia.org/wiki/Supreme_Court_of_India