CAGE


By – Yash Verma, BA. LLB(4th year), KR Mangalam University

Introduction

Imagine a scenario where a rape case that would typically take five to ten years to reach
judgment is instead cleared within one to two. Imagine court processes that once took days
being completed within minutes, or the voluminous drafting and clerical work of a typical
Indian court being executed with the utmost dispatch. Imagine judges assisted by technology
in delivering judgments, and evidence evaluated with the help of technology for rapid
disposal. This is no longer manifestation — it is fast becoming fact. AI, once confined to
being a daily work assistant, is now gaining a foothold within the judiciary itself. India’s
courts have already begun diffusing AI into the justice system to ensure faster justice and
fewer backlogs.

Why AI in Courts

Traditional court processes ensure fair justice but consume voluminous time for case
disposal. Yet justice demands less time. It took 7 years for the Nirbhaya rape case to be
disposed of, 1 15+ years for the Bilkis Bano case, 2 and 14 years for the Priyadarshini Mattoo
case. 3 The timeline is so long because of the voluminous judicial backlog present in the
system. On average, it takes approximately 3–6 years for a case to be disposed of in India. 4
AI is today’s rapid-force technology, solving problems from the simplest to the most complex
within a few minutes. Given the demanding need to save time in the pursuit of justice, AI is being embedded into the judicial system as an assistant that can clear large amounts of
clerical work within minutes.

Multitasker: AI’s Expanding Role

AI’s role is shifting dramatically — what started as just a clerical assistant is becoming a full-
time collaborator, recommending choices, evaluating evidence, and drafting content.
The question that arises is who gets what, and who is affected. AI not just affects judicial
actors in court but also the litigants. Where AI acts as an assistant for the daily workflow of
judicial actors, it has also emerged as a major helper for self-representing litigants. People
who cannot afford legal representation rely on AI for translating legalese into plain language,
navigating procedure, and refining arguments, among other things. AI in the judicial system
is also used as a real-time transcriber for judges’ statements and orders, reducing the
workload of stenography. It is already assisting judicial actors with summarising and drafting,
tasks that take up most of the time in a case. It can also be used to filter and arrange cases
according to priority , as with VICTOR AI of Brazil. APOIA, another AI tool from Brazil,
can draft reports, summarise cases, and find relevant case law — tasks that take up most of
the time during a hearing. 5

AI Diffusion in Indian Courts

Indian courts have already started diffusing AI into the judiciary at a rapid pace, ensuring
complete compliance with guidelines.

  1. SUVAS AI: Launched by then-Chief Justice of India Sharad Arvind Bobde on 26
    November 2019. 6 The tool ensures linguistic accessibility of court orders and
    judgments. Acting as a translation tool, it translates judgments and judicial orders into
    regional languages, helping dismantle language barriers using natural language
    processing. The system ensures large-scale translation with judicial oversight and
    supports open justice.
  2. Adalat AI: A platform that builds AI solutions to help with judicial automation. 7 It
    has recently built a voice AI — a real-time voice-to-text tool designed to meet the
    shortage of stenographers and reduce the workload of depositions taken by hand.
  3. SUPACE — launched on 6 April 2021 by then-Chief Justice S.A. Bobde, in the
    presence of Justice L. Nageswara Rao. 8 It is a research tool developed by the
    Supreme Court’s own team that acts as an assistant, helping judges surf through case
    facts and relevant case law. It strictly performs research and summarisation, not
    adjudication.

What could go wrong

While AI handles complex multitasking with ease, effortlessly outpacing human capability, it
is still shadowed by critical uncertainties and real-world “what ifs.” The question that arises
is: what kind of what-ifs? What could go so wrong with technology this advanced? For
instance, think of a scenario where evidence is evaluated incorrectly due to a systemic error,
or a precedent is hallucinated, or a presiding judge is too lenient on an AI recommendation
and rules out an unjust judgment. There are multiple other harms that could arise due to the
misconduct of AI.

  1. Hallucination: AI is trained on large language models (LLMs). Within LLMs, AI
    works as a predictive agent, predicting what is most likely to be the answer rather than
    the exact and correct one — it gives a probabilistic output. Therefore, in an attempt to
    satisfy the query, AI can sometimes hallucinate. AI could fabricate precedents and
    statutes that don’t exist, directly influencing the court’s decision if left unverified. In
    Deepak v. Heart & Soul Entertainment Ltd. (January 2026), a litigant used AI-
    fabricated citations, for which the Bombay High Court imposed a fine of ₹50,000. 9
  2. Automation Bias: A “frequency of bias” study — ProPublica’s COMPAS analysis 10 showed actual bias rates in the COMPAS risk-assessment algorithm used in the United States. It found that the system classified Black defendants as high-risk while underestimating risk for White defendants. Bias in AI can be deeply rooted from the development stage itself — AI is trained on historical data containing economic, political, cultural, geographical, and social information. This historical data may contain biases such as discrimination and under-representation, which directly influence the biased output of AI. AI in the judiciary, vulnerable to such risk, might produce biased output that could directly influence a biased decision if it remains unverified through human supervision.
  3. Opaque Nature: Courts remain transparent throughout every procedure, from the hearing all the way to disposal, publicising how decisions are reached and what procedures are used to arrive at an outcome. Since the nature of AI is a “black box,” the inner processes and evaluations behind a decision are not visible and may contain uncertainties, errors, or bias. In judicial use, there is often no way to audit why the system produced a given output. This directly conflicts with natural justice principles.
  4. Data Privacy Risks: AI is trained on enormous volumes of data collected across the world from different communities. This raises the question — what if sensitive and judicial records of the courtroom are also collected, or if confidential litigant data entered into the system is used without informing the litigant? If such sensitive data is fed into public AI tools like ChatGPT, it may lead to data leaks. The Kerala High Court’s 2025 policy explicitly warns judicial officers against feeding case data into public AI tools. 11
  5. Accountability Gap: This raises the question of who is responsible, and to whom. One of the core principles of the judiciary is accountability. While AI may ease the workload in judiciary functions, it can still lead to uncertainties and wrong decisions. If any party is impacted by such a decision, who should be held liable — the developer of the tool, or the judicial actor who used it to reach the decision? With all these challenges and risks, the question that arises is: how can AI be safely used, and how are Indian courts adapting to use AI in a safe environment?

India’s Approach towards Challenges

India’s judiciary is currently patching gaps through individual High Court circulars, case-by- case judicial orders, and professional-responsibility rules, rather than any single national law.
High Court Circulars

  • Kerala High Court (July 2025) — Issued India’s first formal AI policy for the
    district judiciary, barring judicial officers from using AI tools to arrive at findings or
    draft orders, and warning against feeding case data into public tools like ChatGPT.
  • Gujarat High Court (April 2026) — Issued its own comprehensive policy restricting
    AI in judicial decision-making.
  • Punjab and Haryana High Court (April 2026) — Issued a circular barring all
    judicial officers from using AI tools — specifically naming ChatGPT, Gemini, and
    Copilot — for legal research or judgment writing.

Liability – Who is responsible

The Bar Council of India has stated that under the Advocates Act, 1961, AI has no legal
recognition, meaning only the human lawyer can be held responsible for erroneous AI-
generated content filed in court. 12
Before any comprehensive and uniform legislative framework, the Indian judiciary across
states is undertaking the necessary steps to mitigate and avoid possible risks. But avoiding
risks does not mean adopting a safe environment for the use of AI — instead, it would only
delay the problem. So how is the Indian judiciary shaping the infrastructure suitable for AI
going forward?

The Catch Up: India’s First Uniform Legislative Framework

The Indian judiciary answers this hurdle by drafting India’s first national legislative framework for AI — the Draft Regulations for the Use of AI in Indian Courts, 2026. 13

About the Framework
Published on 3 June 2026 by the Supreme Court’s AI Committee, chaired by Justice P.S.
Narasimha, it is India’s first comprehensive and uniform legislative framework for AI in the
judiciary — not just for the Supreme Court, but binding-in-intent across the entire
adjudicatory ecosystem.
Proposed, Not Enacted
The framework is still under consultation and not yet enacted. It remains open for revision
based on feedback from judges, lawyers, technology companies, and civil society.


Core Principles

  1. Human Primacy — The framework establishes the independence and primacy of
    human decision-making, stating that AI remains subordinate to human oversight. It
    mandates the crucial “human-in-the-loop” principle, ensuring final decisions are
    taken by humans.
  2. Transparency — Mandates disclosure of AI use — litigants or judicial actors using
    AI for drafting must explicitly disclose it.
  3. Accountability — Responsibility and liability for any wrong decision or incorrect
    output rests solely on the human user.
  4. Judicial Independence — Mandates that the decisions of judicial actors and judges
    must remain solely independent. They must not be unduly influenced by AI.
  5. Data Safety — Mandates necessary steps to ensure the safety of sensitive courtroom
    data and confidential information.
    Permissible Areas
    AI may assist with a broadly defined range of tasks: research, drafting assistance, scheduling
    of hearings, transcription, translation of judgments, citation verification, case management,
    and record/judicial administration.

Prohibited Areas

  • Deciding cases
  • Determining bail eligibility
  • Assessing the credibility of witnesses or parties
  • Influencing judicial deliberations
  • Monitoring judicial officers, litigants, or advocates
  • Risk scoring

    Institutional Body for the Framework
    A permanent, full-time “Apex Body” is proposed to sit at the Supreme Court, tasked with
    setting standards, approving AI systems before court deployment, and overseeing
    compliance. Its composition is broad and cross-institutional: two Supreme Court judges, two
    High Court Chief Justices, two High Court judges, a member of a reputable institution, an
    officer not below Joint Secretary rank from the Ministry of Electronics and Information
    Technology, and subject experts in finance, cybersecurity, data privacy, technology law, and
    AI. Beneath it sit the AI Committee itself, an AI Secretariat, and CoRE-AI (the Centre of
    Research and Excellence on AI).

Connecting Dots with Global Framework

While India is writing its first uniform legislative framework, other jurisdictions appear to be
following the same path — some behind, some ahead.
Brazil
In Brazil, the National Council of Justice issued a binding, 12-chapter framework
(Resolution 615/2025) 14 that governs how AI tools are designed, procured, deployed, and
audited across the entire Brazilian judiciary. It works by classifying every AI system into risk
categories, requiring mandatory audits and impact assessments before deployment, and
mandating explainability so that any decision can be traced and understood rather than treated
as a black box.

The framework establishes a dedicated National Committee for AI in the Judiciary —
drawing from judges, bar associations, public defenders, and civil society — which oversees
compliance and publishes annual public evaluation reports.
USA
The US has no binding federal framework comparable to India’s draft as of now, aside from
state-regulated laws that have led to a fragmented, inconsistent patchwork. A 2026 survey
found that 61.6% of federal judges already use some AI tool in their work, yet only 9%
believed generative AI should be used in judicial work at all. 15 India’s decision to regulate
before AI becomes routine in chambers is arguably the more prudent sequencing.
Europe
The EU AI Act offers the clearest structural approach: a risk-tiered system where AI used in
the administration of justice is classified as “high-risk,” 16 triggering mandatory conformity
assessments, human oversight, and transparency obligations before deployment —
conceptually close to India’s proposed pre-approval, Apex Body-certified model, though the
EU framework carries binding legal force that India’s draft does not yet have.

Indian Framework as an Influence for other Jurisdiction

India’s real value as a model is timing: it drafted rules while its hallucination crisis was still
unfolding, rather than reacting after the fact, as Brazil and the US did. Three exportable
takeaways stand out:

  1. Its unambiguous ban on AI risk-scoring, offering a cleaner red line than more
    narrowly worded restrictions elsewhere.
  2. Its low-cost lawyer disclosure requirement, which smaller judiciaries could adopt
    immediately without needing a full institutional apparatus.
  3. Its single, unified framework covering both administrative and adjudicatory AI use.

Where does India Stand

The Indian judiciary already has AI tools in use that carry considerable risk and can produce
real harm. To tackle these risks, courts are continuously patching the gaps with circulars and
guidelines in the absence of a uniform legislative framework.
But a rulebook is already being drafted. India’s judiciary has moved from reacting to
systematising. The tools already in use — SUPACE, SUVAS’s translated judgments, Adalat
AI — show a judiciary embracing AI’s efficiency gains well before it had rules to govern
them. The Draft Regulations, 2026 mark the shift from a reactive patchwork to an actual
system: a single Apex Body, a non-negotiable ban on risk-scoring, and a disclosure regime
meant to close the very gap the earlier sections exposed. Set against Brazil’s year of binding
enforcement and the United States’ unfinished tangle of pending bills, India’s attempt stands
out precisely because it was drafted while the crisis was still live.
The Supreme Court has already told us what happens when a judgment rests on something
that was never real. But a system only earns that name once it’s tested. Until the Apex Body
actually certifies a tool, audits a failure, or disciplines a violation, India has completed the
easier half of the journey — writing the rules — and not yet the harder half: proving they
hold.

Leave a Reply