Policy on Use of Artificial Intelligence in Judicial and Court Administration (High Court of Gujarat, 2026)

Gujarat's policy followed Kerala's July 2025 policy for district courts,⁵ and arrived in the same season as a wave of fabricated-citation incidents. The Supreme Court had declared in February 2026 that a trial court's reliance on AI-generated, non-existent judgments is misconduct,⁸ and in March 2026 remarked that citing non-existent AI-generated judgments was rampant in courts in India and abroad.⁹

Bar & Bench reported on 4 April 2026 that the Gujarat High Court "has released a policy barring the use of artificial intelligence tools by judges and court staff in the State for drafting orders, judgment preparation or any other form of judicial decision-making".¹ The report does not give the date of the policy's issue, and the policy document itself was not available for this entry, so the description below relies on that report.

Foundations and scope. The policy is framed under Articles 225 and 227 of the Constitution, which govern the High Court's jurisdiction and its superintendence over subordinate courts, and rests on the right to a fair hearing under Article 21. It applies to all judicial officers, court staff, legal assistants, interns and para-legal volunteers engaged with the High Court of Gujarat and the district judiciary.¹

The prohibition. The policy declares that AI "shall never be employed for any form of decision-making, judicial reasoning, order drafting, judgment preparation, bail/sentencing considerations, or any substantive adjudicatory process". It expressly prohibits using AI to arrive at, determine or substantially influence any finding of fact, finding of law or operative order, "even if such output is subsequently reviewed by a judge", and bars its use for sorting, classifying or organising evidence, assessing credibility, filtering relevance, summarising depositions or testimony, or any task involving evaluation or categorisation of proof.¹

Permitted assistive uses. The policy allows AI for legal research, retrieval and analysis of judgments, extraction of ratio and identification of precedents, provided results are confirmed against approved law reports; for administrative tasks such as code generation, IT automation, internal training templates and drafting circulars and notices; and for drafting assistance that improves the language, structure and clarity of drafts, provided that "the substantive legal analysis and reasoning remains entirely of the judge".¹

The Gujarat policy puts unusual weight on personal liability. A judge's responsibility for every order, judgment and observation issued in his or her name "cannot be delegated, shared with, or diminished by the use of any AI tool". Any AI-generated output, "once signed or authenticated by a user, becomes the sole responsibility of that user", and the signatory is liable for any inaccuracy or omission. "The use of AI does not constitute a defence to a finding of error, misconduct, or professional negligence. Users cannot disclaim responsibility by attributing errors to an AI tool."¹

Citations generated by AI, and case references or statutory provisions, cannot be used without independent verification from authoritative primary sources, which the report lists as including AIJEL, SCC Online, AIR, the Supreme Court website and official gazettes. Legal assistants, research associates and judicial assistants who use AI to help a judge must make sure the judge knows, and where AI is used in a research note, bench memo or legal brief, the responsible officer must record the fact.¹

The policy bars entering personal and sensitive information into public AI tools, including names, addresses or identifying details of parties, witnesses or advocates, details of pending proceedings or unreported orders, privileged communications and confidential legal strategy, sensitive personal data such as health, financial, biometric or caste-related information, and evidence or documents filed in a case. It warns that AI may encode or perpetuate bias on grounds including gender, religion, caste, ethnicity or socio-economic status. Violations are treated as misconduct and attract departmental or disciplinary proceedings under service rules, in addition to any civil or criminal liability under the Information Technology Act, 2000 or the Bharatiya Nyaya Sanhita, 2023. The policy is "subject to any direction or circular issued by the Supreme Court of India or its e-Committee regarding AI use in courts, or any policy issued by the legislature".¹

A reported enforcement of the High Court's AI norms reached outside the judiciary. Faiz Enterprise, a Surat proprietorship, challenged a GST registration cancellation and the orders rejecting revocation and appeal (Special Civil Application No. 10123 of 2026). On 13 August 2026 its counsel argued that the State Tax Officer had cited case laws that were non-existent or inapplicable, among them State of Gujarat v. Aarbee Structures Pvt. Ltd. (2024) and decisions attributed to the Madras and Calcutta High Courts. The Bench of Justices A.S. Supehia and Vaibhavi D. Nanavati observed that the impugned order appeared to rely exclusively on AI-generated case laws and directed the Government Pleader to take instructions.²

The officer, Devang Arvindkumar Yadav, filed an affidavit tendering an unconditional apology, saying he was a probationary officer, that he had used AI to draft the order and that he had cited non-existent and irrelevant judgments, and that he had since attended a training programme on the use of AI in drafting and adjudication. On 18 August 2026 the Additional Commissioner of State Tax issued statewide instructions: AI may only assist; any judgment, statute, rule, circular or notification identified through AI must be independently verified from primary official sources; if AI paraphrases a ratio, the officer must find and reproduce the original paragraph, check that the judgment is still good law and record reasons if a taxpayer's authorities are held inapplicable; AI output "may assist in reasoning but cannot substitute it"; the issuing authority remains responsible; and violation is misconduct that may attract disciplinary action.²³

In its judgment of 20 August 2026 the High Court directed that the instructions "be scrupulously followed" and that "any violation of the instructions would amount to contempt of this Court". It quashed the show cause notice, cancellation order, revocation-rejection order and appellate order, which the State itself proposed to revisit, and directed fresh notice and a reasoned order.²

The four main instruments now in circulation differ in scope and strictness.¹⁵⁶

The Punjab and Haryana instruction was communicated on 6 April 2026 by the High Court administration in a letter to the District and Sessions Judges of Punjab, Haryana and Chandigarh. It says the Chief Justice has asked that judicial officers be directed "not to use Artificial Intelligence tools including but not limited to Chat GPT, Gemini, Copilot, Meta etc. for writing of judgements and legal research. Any violation of these instructions will be viewed seriously".⁴⁵⁶

Gujarat's policy answers one criticism made of Kerala's: it applies to the High Court's own judges and staff, so its standards are not limited to trial courts.¹⁵ It also goes further than the others in barring AI from evidence-sorting and summarisation, tasks such as summarisation that the Supreme Court's draft Regulations would permit with approval and verification, and it relies on personal liability rather than approved-tool lists. Because it yields expressly to Supreme Court or e-Committee directions, its long-term force depends on whether the national draft is adopted. The sources reviewed contained no published commentary on the policy beyond news reports, and the policy text itself was not available for this entry.

The Faiz Enterprise case shows how the principles travel to quasi-judicial authorities that decide tax and regulatory disputes. The Supreme Court's 2 September 2026 judgment in Vijay Ghanshyam Gadiya v. Union of India, which set aside a customs penalty for reliance on AI-hallucinated case law, addressed the same kind of authority.⁷

As of 15 October 2026 the Gujarat policy is the operative AI instrument for the High Court and the State's district judiciary, and the statewide State Tax instructions of 18 August 2026 are enforceable through the High Court's contempt warning in Faiz Enterprise. The sources reviewed did not show amendments, reported disciplinary action under the policy, or the High Court's response to the Supreme Court's draft regulations.