Keeping the ‘judge’ in judgment

‘First published on India Business Law Journal’

By: Pravin Anand, Ajai Garg and Siddhant Chamola

The Supreme Court of India’s Artificial Intelligence (AI) Committee has put out India’s first structured rules for AI in the courtroom and they make human primacy paramount. The rules are currently in draft form, pending finalisation based on comments from various sectors.

Overview of India’s AI regulations

On 3 June 2026, the above-mentioned AI committee published a preliminary draft, the Regulations for Use of Artificial Intelligence in Courts, 2026, and invited comments from the public.

The draft is among the first structured attempts in India to set terms for how the judiciary and other associated bodies may adopt AI across judicial and administrative work. Its core principle is human primacy: AI must play a strictly assistive function and the power to take decisions must rest with judges alone.

The draft prohibits: the determination of outcomes through algorithmic decision making; risk scoring for bail or recidivism; and the profiling of parties and witnesses. These provisions draw a clear line between assistive intelligence, which supports judges, and substitutive intelligence.

The draft also confronts the “hallucination” issue, which has seen a disturbing rise, not just in India, but in several judicial systems across the world. Earlier, the world had witnessed legal briefs and research submitted by lawyers with fake citations of cases cooked up by AI. Recently, we have come across judgments overturned over the revelation that they, too, had cited hallucinated cases.

The draft regulations define “hallucination” with precision, and advise that AI outputs be treated as advisory, requiring thorough verification for accuracy before they are relied on.

The draft goes further by asking lawyers to file a signed declaration disclosing any AI assistance in preparing a pleading or document, and it bars anyone from later pleading the AI character of the output as a defence.

Each pleading, submission, order and judgment carries tremendous responsibility not just for the case it addresses, but for future cases. The requirement to provide an undertaking is a welcome step that will instil accountability and reduce unverified and irresponsible AI usage.

Suggestions to strengthen draft AI

Below are suggestions offered in constructive spirit. None quarrels with the architecture of the draft.

First, verification should extend to legal research and citation retrieval. The draft permits AI-assisted transcription and translation only with mandatory human verification, yet the clause covering legal research, precedent retrieval and citation verification carries no matching requirement. Since fabricated citations are the most common way AI has gone wrong in courtrooms, an express verification obligation in that clause would close the most obvious gap in the permissible uses regime.

Second, two definitions could be sharpened. The draft defines a “black box” by tying it to deep learning, yet opacity is not the property of any one technique. Anchoring the definition to the absence of explainability, rather than to a named method, would keep it durable as the technology evolves.

Separately, the term “high-risk AI tools” appears in the procurement chapter without a definition. A simple tiering would assist everyone: tools that help draft orders at the high end; scheduling and case management tools in the middle; and translation or accessibility tools at the low end.

Third, confidentiality should reach beyond personal data. The draft applies its data minimisation principle to personal data, following the Digital Personal Data Protection Act. Commercial information, trade secrets and material covered by non-disclosure obligations sit outside that act. Where such information enters the court record (and in commercial litigation it routinely does), a carve-out treating it as a distinct protected category would guard against inadvertent exposure through AI processing.

The committee has produced a document that is thoughtful and, in its instincts, sound. It reflects an institution willing to adopt useful technology, while remaining firm on the things that make a court a court. The implementation of the regulations for AI in the judiciary will be very welcome, once all comments are assessed and the draft is finalised.