Two Courts, Two Philosophies
When the Rhode Island Supreme Court adopted interim guidelines for generative AI on June 18, 2026 — amending Rule 1.1 to require lawyers to understand "the benefits and risks associated with existing and developing technology" and warning that AI "must not undermine or replace a judge's independent decision-making" — it was, by design, a light touch: advisory, short, and grafted onto an existing competence rule.
India had already gone further. On June 3, 2026, the Supreme Court's AI Committee released the Draft Regulations for Use of Artificial Intelligence in Courts, 2026, inviting public comment through June 20. Unlike Rhode Island's incremental amendment, this is a from-scratch, nationally binding framework meant to cover the Supreme Court, every High Court, and every tribunal and statutory adjudicatory body in the country — an architecture, not a footnote. Both instruments land on the same first principle. India's draft calls it "human primacy and judicial independence": AI can assist, but "judges will have the full power to decide every case," as The Leaflet's explainer summarizes it. Where the two diverge is in how much apparatus that principle is asked to carry.
The Case the Regulators Have Earned
The strongest argument for India's more elaborate approach is that it isn't hypothetical. Indian courts have already lived through the failure mode these rules exist to prevent. A trial court in Karnataka relied on non-existent Supreme Court rulings generated by ChatGPT; Justice B.V. Nagarathna encountered a fabricated citation styled "Mercy v. Mankind" during a PIL hearing; and in January 2026 the Bombay High Court imposed ₹50,000 in costs on a litigant who filed AI-invented case law, according to LiveLaw's reporting on "phantom precedents." A judiciary carrying roughly 50 million pending cases, under pressure to adopt any tool that promises speed, has genuine reason to worry that hallucinated citations could slip past overworked benches and into final orders. Kerala's High Court got there first institutionally: its July 19, 2025 policy for the district judiciary barred AI from being used to "arrive at any findings, reliefs, orders, or judgments under any circumstances," per S.S. Rana & Co.'s summary — nearly eleven months before the Supreme Court's national draft caught up. Given that track record, a rule demanding disclosure, human verification, and a ban on outcome-determinative algorithms is not regulatory overreach. It is triage.
Where the Draft Overreaches
The overreach lies not in the prohibitions but in the permissions. India's draft bans the right things absolutely — risk-scoring, profiling litigants, and treating AI output as independent evidence are all non-derogable, per The Leaflet's reading of the text. But it also requires prior written approval before courts can deploy AI for case management, transcription, translation, or legal research — the low-risk, high-value uses that Indian courts have already been running for years through SUPACE and SUVAS, the Supreme Court's own research and 19-language translation systems. Layered on top is a five-committee Apex Body — Judicial, Technical, Infrastructure/Finance, Case/Data Management, and Cyber Security — with every High Court expected to stand up its own AI Committee and secretariat. That is a heavier institutional footprint than the harm being addressed strictly requires, and it invites the outcome regulators claim to want to avoid: courts starved for capacity now need a written approval workflow before they can turn on a translation tool that already works.
Rhode Island's model separates the two problems more cleanly. It targets the actual failure — a lawyer or judge trusting unverified output — with a competence duty and a verification obligation, and leaves the tooling decision to the professional using it, backed by malpractice and disciplinary exposure if they get it wrong. That is arguably a lighter, more durable design: it scales with whatever AI tools exist in five years without a legislature-style body re-approving each one.
What India Should Take From Its Own Experiment
"AI can be used for assistance, whereas decisions about law, fact and justice belong entirely to the court."
That line, translated from India's Regulation 4, is exactly right, and it should survive the June 20 consultation untouched. What deserves revision is the assumption that every assistive use needs the same pre-clearance as an outcome-affecting one. A workable final rule would keep the absolute bans, keep mandatory disclosure to litigants when AI materially shapes a proceeding, and keep personal accountability on the judicial officer who relies on an AI output — but replace blanket prior approval for low-risk categories (translation, transcription, scheduling) with a lighter registration-and-audit regime, reserving the Apex Body's bandwidth for genuinely novel or higher-risk deployments. India moved first and moved seriously; the test now is whether the Committee can trim the bureaucracy without touching the principle that got this right in the first place.