GUARDIAN OF THE GAVEL: A CRITICAL ANALYSIS OF THE GUJARAT HIGH COURT'S POLICY ON THE USE OF ARTIFICIAL INTELLIGENCE IN JUDICIAL AND COURT ADMINISTRATION

CONSTITUTIONAL FOUNDATIONS AND LEGISLATIVE FRAMEWORK

The policy derives its authority from two principal constitutional provisions. Article 225 preserves the jurisdiction of existing High Courts, while Article 227 confers upon the High Court the power of superintendence over all courts and tribunals within its territorial jurisdiction. The invocation of these provisions is jurisprudentially sound. The power of superintendence is plenary in nature and has been consistently construed broadly by the Supreme Court of India to include administrative as well as judicial oversight.

The policy further situates itself within a constellation of statutory instruments, namely the Information Technology Act 2000, the Digital Personal Data Protection Act 2023, the Contempt of Courts Act 1971, and the High Court of Gujarat Rules 1993. The DPDP Act reference is particularly prescient, as the data privacy dimensions of AI use in courts represent one of the most pressing regulatory challenges of the present era.

The anchoring of the policy in the right to a fair hearing under Article 21 is its most significant constitutional gesture. It affirms that the procedural guarantees embedded in the right to life and personal liberty are not merely negative obligations upon the state but carry positive institutional duties, including the duty to ensure that adjudication remains human, reasoned, and accountable.

THE PHILOSOPHICAL UNDERPINNING: HUMAN CONSCIENCE AS THE SINE QUA NON OF ADJUDICATION

The policy is not merely a regulatory instrument. It is, at its core, a philosophical declaration. Clause 1.2 articulates with clarity that the weighing of evidence, interpretation of law, application of legal principles to facts, exercise of discretion, and delivery of reasoned decisions belongs exclusively to the domain of the human mind.

This articulation draws upon a deep jurisprudential tradition. Justice, in the constitutional sense, is not the mere application of rules to facts. It is the exercise of conscience informed by law, reason, and the lived realities of those who come before the court. An algorithm, however sophisticated, operates upon statistical patterns derived from historical data. It cannot exercise conscience. It cannot weigh the dignity of the individual against the imperatives of the state with the empathy that Article 21 demands.

The policy also identifies a systemic risk that is often understated in technology debates: gradual over reliance. The concern is not merely that a judge may one day issue an entirely AI authored judgment. The more insidious risk is incremental dependence, where over successive uses, the judge's independent faculty of reasoning is quietly supplanted by algorithmic suggestion. The policy addresses this risk with institutional discipline by confining AI to the narrowest conceivable role.

SCOPE AND APPLICATION: A COMPREHENSIVE AMBIT

Clause 5 of the policy deserves particular commendation for its breadth of application. The policy governs not merely judges but all persons within the judicial and administrative framework, including legal assistants, interns, trainees, para-legal volunteers, and contractual staff. This is a deliberate and necessary choice. In practice, AI tools are more likely to be used by junior research assistants preparing bench memos than by judges directly. By extending liability to every person in the judicial ecosystem, the policy forecloses the evasion of responsibility through delegation.

The coverage of devices and platforms under Clause 5.3 is equally comprehensive. The policy applies regardless of whether AI tools are accessed via court owned hardware, personal devices used for court related work, third party platforms, mobile applications, browser extensions, or API based integrations. This is a significant drafting achievement. Many institutional policies restrict themselves to official devices and thereby create enforcement blind spots. The Gujarat policy expressly closes those gaps.

The enumeration of covered AI tools in Clause 5.4 is notable for its specificity. It identifies by name large language models such as ChatGPT, Gemini, Copilot, DeepSeek, Claude, and Grok, while also covering AI assisted legal research platforms, machine translation tools, transcription systems, document summarisation tools, predictive analytics platforms, and AI functionality embedded in productivity software. This taxonomy reflects a sophisticated understanding of how AI actually operates within institutional workflows.

PROHIBITED USES: THE INVIOLABLE CORE

Clause 8 of the policy sets out absolute prohibitions and merits careful examination. The prohibition is comprehensive in its reach. Artificial Intelligence shall never be employed for any form of decision making, judicial reasoning, order drafting, judgment preparation, bail or sentencing considerations, or any substantive adjudicatory process. The use of the word never is constitutionally and institutionally deliberate. It admits of no exception, no qualification, and no progressive relaxation.

Three specific prohibitions stand out as particularly significant.

First, the prohibition on using AI for evidence handling, including sorting, classifying, organising, assessing credibility, filtering relevance, and summarising depositions or testimony, addresses a danger that has received inadequate attention in public discourse. The evaluation of evidence is not a mechanical exercise. It requires the judge to observe witness demeanour, to weigh contradictions, and to apply the standard of proof with human judgment. Any AI involvement in this domain, even ostensibly ministerial, risks contaminating the factual foundation of the adjudicatory process.

Second, Clause 8(5) enumerates with specificity the categories of information that shall never be entered into any public AI tool. These include names and identifying information of parties and witnesses, details of pending proceedings and unreported orders, privileged communications, sensitive personal data including health, financial, biometric, and caste related information, and evidence or documents filed in a case. The inclusion of caste related information is particularly contextually sensitive and reflects an awareness of the social dimensions of data protection in India.

Third, Clause 8(7) prohibits reliance on AI generated citations without independent verification from authoritative primary sources such as AIJEL, SCC Online, AIR, the Supreme Court website, or official government gazettes. This provision directly responds to the documented phenomenon of AI hallucination, whereby language models generate plausible but entirely fictitious case citations. Courts across India, and indeed globally, have encountered instances of fabricated precedents finding their way into pleadings and even judicial orders. This prohibition, coupled with the verification obligations in Clause 10, constitutes the policy's most operationally significant safeguard.

PERMITTED USES: A CALIBRATED CONCESSION TO EFFICIENCY

The policy is not a wholesale rejection of technology. Clause 7 delineates permitted uses with equal precision. AI may be employed for legal research and retrieval of judgments, extraction of ratio decidendi, identification of precedents, and statutory interpretation, provided all outputs are verified by applying an independent human mind and confirmed against approved law journals. It may be used for drafting assistance to improve language, structure, and clarity of draft orders, provided the substantive legal analysis and reasoning remain entirely that of the judge. It may assist with machine translation, transcription, case scheduling based on anonymised metadata, and administrative reporting.

The internal logic of these permissions is sound. They permit AI to function as a research accelerator and administrative rationaliser without permitting it to penetrate the domain of judicial reasoning. The distinction between a tool that surfaces potentially relevant precedents and a tool that determines which precedent applies is precisely the distinction the policy seeks to preserve.

PERSONAL LIABILITY: ACCOUNTABILITY AS A CONSTITUTIONAL IMPERATIVE

Perhaps the most legally consequential provision of the policy is Clause 11, which establishes a regime of personal and non delegable accountability. Every judge is personally responsible for every order, judgment, and observation issued under their name. This responsibility cannot be delegated, shared with, or diminished by the use of any AI tool. The use of AI does not constitute a defence to a finding of error, misconduct, or professional negligence.

This provision has profound implications. It forecloses the emerging tendency, already visible in some jurisdictions, of attributing judicial errors to AI systems rather than to the human officers who relied upon them. It restores and reinforces the constitutional principle that every judicial act is a personal act, bearing the imprimatur of individual conscience and individual accountability.

The extension of this principle to legal assistants and research associates under Clause 11(3) is equally important. These officers are required to ensure that the concerned judge is informed of any AI use and of any AI assisted output. Transparency within the judicial hierarchy is made a formal obligation, not merely a professional courtesy.

CRITICAL OBSERVATIONS

A candid analysis requires acknowledgment of certain limitations and tensions within the policy.

The policy does not establish any institutional mechanism for enforcement or audit. Clause 12 provides that violations shall be treated as misconduct attracting departmental or disciplinary proceedings, but no provision is made for monitoring compliance, detecting undisclosed AI use, or auditing outputs for signs of AI generation. In the absence of a detection and reporting framework, the policy risks operating as a declaratory instrument rather than an enforceable regulatory regime.

Further, the policy restricts AI use to tools approved by the High Court, but no list of approved tools is appended or referenced. Practitioners and officers are left without clear guidance on which platforms meet the institutional standard for enterprise or private deployment under Clause 9.3. This lacuna invites inconsistency and requires urgent resolution by way of supplementary notification.

The policy also does not address the position of advocates and litigants. An advocate who prepares arguments using AI tools and presents fabricated citations before the court may not fall within the personal scope of Clause 5.1. While the Supreme Court has issued general warnings on this issue and professional conduct rules may apply, the absence of a specific provision within this policy leaves a significant regulatory gap.

Finally, the review mechanism under Clause 13, while appropriately preserving the High Court's authority to revise the policy in light of technological developments or Supreme Court directives, does not specify a review timeline. Given the pace of AI development, a periodic mandatory review, perhaps annual, would better serve the policy's stated objective of providing a basis for regular review and evolution of AI governance.

COMPARATIVE CONTEXT

The Gujarat policy does not exist in isolation. Kerala High Court had, in July 2025, barred district court judges from using AI to arrive at findings, reliefs, orders, or judgments. The Punjab and Haryana High Court followed Gujarat's approach within days by issuing an administrative circular to all district and sessions judges proscribing the use of named AI tools for judgments and legal research. The Supreme Court of India has itself described AI generated fictitious citations as a growing menace. Internationally, courts in the United States have imposed sanctions on attorneys who filed AI fabricated precedents, and several jurisdictions have introduced mandatory AI disclosure rules for court filings.

The Gujarat policy is distinguishable from these comparative examples by the comprehensiveness of its framework. Where other courts have issued warnings or brief circulars, the Gujarat High Court has produced a twelve clause constitutional and statutory instrument complete with definitions, guiding principles, permitted and prohibited uses, data protection obligations, verification requirements, accountability standards, and consequences of violation. It is, in the taxonomy of AI governance instruments, a model policy.

CONCLUSION

The Policy on Use of Artificial Intelligence in Judicial and Court Administration issued by the Gujarat High Court is a jurisprudentially grounded, institutionally disciplined, and constitutionally anchored instrument. It correctly identifies the inviolable core of adjudication as a domain of human conscience that no algorithm can legitimately enter. It draws a legally defensible line between AI as a servant of efficiency and AI as a usurper of judgment.

Its limitations, principally the absence of an enforcement mechanism, the lack of an approved tools registry, its silence on the position of advocates, and the absence of a mandatory review timeline, are capable of remedy by supplementary notification and do not detract from the fundamental soundness of its architecture.

For the Indian judiciary, which bears the weight of tens of millions of pending cases and the constitutional trust of over a billion citizens, this policy charts a responsible course: harness what technology can offer without surrendering what justice demands. The gavel must remain in human hands. The Gujarat High Court has said so, and said it well.

Link : https://gujarathighcourt.nic.in/hccms/sites/default/files/miscnotifications/Policy%20on%20the%20use%20of%20Artificial%20Intelligence%20in%20the%20Judicial%20and%20Court%20Administration.pdf