The clause in dispute
On July 19, 2026, the Supreme Court Advocates-on-Record Association (SCAORA) filed formal comments opposing one specific provision in the Supreme Court's Draft Regulations for Use of Artificial Intelligence in Courts, 2026 — a framework the Court's AI Committee released for public consultation on June 3, 2026. Regulations 43(3) and 43(4) of the draft would require advocates to disclose, through a formal certificate, whenever an AI tool was used in preparing a pleading, document, or piece of evidence. SCAORA called the requirement "unworkable," arguing that advocates are already personally responsible for the accuracy of every filing under the Advocates Act, 1961 and the Supreme Court Rules, 2013 — a disciplinary regime, enforced through Section 35 of the Act, that has governed professional misconduct for 65 years. In SCAORA's words, "the ultimate filing remains the sole responsibility of that advocate, who personally vouches for the accuracy of its contents."
It's worth being precise about what the draft does not touch. The regulations otherwise permit AI-assisted legal research, drafting, translation, transcription, and case management across the Supreme Court, High Courts, subordinate courts, and tribunals, while flatly barring AI from deciding cases, assigning bail, or influencing sentencing — output that must remain, in the Committee's phrasing, an exercise of "autonomous human decision making." SCAORA's objection is narrow: it wants the disclosure certificate gone, not the broader framework.
The steelman for disclosure
The case for a disclosure rule isn't frivolous. Generative AI tools hallucinate case citations — a problem that has already produced sanctioned lawyers and struck pleadings in multiple jurisdictions. A mandatory certificate creates a clean paper trail: if a filing turns out to contain a fabricated precedent, the court knows immediately which tool produced it and can trace the failure rather than treating it as an isolated lapse in diligence. It also gives judges — who are not required to be technologists — an explicit signal to scrutinize a document's citations more closely, rather than relying on a general presumption of accuracy that AI-assisted drafting may not deserve. For a judiciary trying to get ahead of a fast-moving technology rather than react to a scandal, a blanket disclosure default is a defensible starting position.
Why SCAORA's narrower fix is the better rule
The weakness in the blanket approach is one of scope and enforceability, not principle. "AI tool" today spans everything from a grammar checker to a fully generative drafting assistant, and the draft regulations, as reported by Supreme Court Observer, don't clearly distinguish between them. A certificate requirement triggered by spell-check-adjacent software isn't a transparency measure; it's a compliance tax that trains lawyers to either over-disclose defensively or quietly under-report, defeating the rule's purpose either way. SCAORA's proposed alternative — amending the existing Special Leave Petition certificate to add a line certifying that the advocate has personally verified citations and precedents — targets the actual harm (uncorroborated AI output reaching a judge) without requiring a compliance filing every time a lawyer opens a research tool. That is proportionate regulation: it enforces the substantive obligation courts actually care about — accuracy — rather than the incidental fact of which software assisted in drafting.
SCAORA's broader submission reinforces that this isn't resistance to oversight. The association also asked for a phased rollout of AI in courts, stronger data-protection safeguards, technical and cybersecurity audits of AI tools already deployed in the judiciary — including SUPACE, SUVAS, and SuSahayak — and guaranteed advocate representation on the AI Committee's governance structure. That's a bar body asking for more institutional accountability, just routed through audits and representation rather than a per-filing certificate that would sit unread in most case files.
The rest of the framework holds up better
Set against the disclosure fight, the draft's core architecture is sound policy. Barring AI from bail and sentencing decisions responds to a real and well-documented risk: opaque risk-scoring tools embedding bias into outcomes that affect liberty. Restricting AI to research, drafting, and administrative functions while keeping final judgment with a human decision-maker is the correct line to draw, and it's one India's Supreme Court reached without needing a scandal to force the issue — a genuinely proactive piece of governance. The layered oversight structure proposed alongside it, running from a Supreme Court apex body down through High Court AI committees to a dedicated AI Secretariat, gives the framework institutional teeth beyond a single disclosure clause.
What should happen next
The AI Committee should adopt SCAORA's substitution: fold AI-verification into the existing certification framework advocates already file, and drop the standalone disclosure mandate. That preserves the accountability the draft is reaching for without creating a parallel compliance regime that duplicates the Advocates Act, invites gaming through vague tool definitions, and risks prejudicing litigants whose lawyers happened to run a citation through a research assistant. The rest of the framework — the bail and sentencing bar, the audit requirements for existing court AI tools, the tiered oversight body — should proceed largely as drafted. Getting the disclosure clause right matters because this framework will likely become the reference model other Indian High Courts and tribunals adapt; a redundant, poorly bounded disclosure rule at the top will propagate downward.