Policy Regarding Use of Artificial Intelligence Tools in District Judiciary (High Court of Kerala, 2025)

By mid-2025 generative AI tools were in everyday use among Indian lawyers and, informally, among judicial officers. In February 2025 a Kerala High Court Division Bench hearing a challenge to the concept of brain death recorded that it "searched online" and that ChatGPT "provided us with the prevalent policies", a chart it placed on record but did not rely on for its reasoning.¹² Courts elsewhere in India had also begun to see AI-drafted material. The Punjab and Haryana High Court had consulted ChatGPT in a 2023 bail order, and trial courts and tribunals were beginning to cite judgments that could not be found.

Against that background the Kerala High Court framed a policy for the lower courts it supervises. A Joint Registrar's Official Memorandum dated 19 July 2025 told all District Judges, the District Judge at Kavaratti and all Chief Judicial Magistrates that the High Court had formulated the policy "for the responsible and restricted use of Artificial Intelligence in judicial functions of the District Judiciary", and asked them to pass it to every judicial officer and staff member and "ensure strict compliance".¹ Bar & Bench reported it the same day as a first in the country.² The Leaflet published the policy document on 22 July 2025.⁴

The introduction acknowledges that AI tools can have a major impact in many fields but warns that their indiscriminate use may cause "violations of privacy rights, data security risks, and erosion of trust in judicial decision-making". Its stated objectives are that AI tools be used only responsibly, "solely as an assistive tool" and "strictly for specifically allowed purposes", and that "under no circumstances" are they used as a substitute for decision-making or legal reasoning.¹

Scope (clause 2). The policy applies to every member of the district judiciary in Kerala and the employees who assist them, and also to interns and law clerks. It covers all kinds of AI tools, including generative AI and databases that use AI to give access to laws and statutes, and it applies wherever and whenever the tools are used, whether on personal devices, court-owned devices or third-party devices.¹

Definitions (clause 3). "Generative AI" is defined as AI that uses large language models to generate text, speech or images, with ChatGPT, Gemini, Copilot and DeepSeek given as examples. "AI tools" include any application that generates text or finds case law or statutory provisions. "Approved AI Tools" are those that have undergone formal evaluation and received approval from the High Court of Kerala or the Supreme Court of India for use in the district judiciary.¹³

Guiding principles (clause 4). Transparency, fairness, accountability and confidentiality "shall not be compromised" by AI (4.1). Because popular tools such as ChatGPT and DeepSeek are cloud-based and may use whatever users enter, including to fine-tune their models, officers must not submit case facts, personal identifiers, privileged communications or litigation documents to them, and the use of all cloud-based services "should be avoided, except for the approved AI tools" (4.2).¹⁵

The policy draws its main line in clause 4.6: AI tools "shall not be used to arrive at any findings, reliefs, order or judgment under any circumstances, as the responsibility for the content and integrity of the judicial order, judgment or any part thereof lies fully with the Judges."¹ SpicyIP describes this as the policy's categorical red line.³ The Supreme Court's 2026 draft regulations, by contrast, adopt a presumption in favour of responsible AI adoption while keeping adjudication in human hands.

Within that limit, approved tools may be used as aids. Clause 4.5 allows them for "routine administrative tasks such as scheduling of cases or court management", always under human supervision, and clause 4.7 limits each tool to the purposes for which it was supplied. SpicyIP's analysis lists legal research, transcription and translation or summarisation of case records among the assistive uses the policy contemplates.³ Where AI is used for translation of legal texts or case law, the output must be verified by qualified translators or the judges themselves (4.4).¹

The policy also reflects a concern about hallucination. Clause 4.3 records that most AI tools produce "erroneous, incomplete, or biased results". For that reason, even for approved tools, any results "including, but not limited to, legal citations or references, must be meticulously verified by the judicial officers", and this expressly extends to databases of case law and statutes that use AI to generate, summarise or fine-tune results.¹

Three procedural duties give the policy its accountability structure. First, audit: under clause 4.8 courts must "maintain a detailed audit of all instances wherein AI tools are used", including the tools used and the human verification process adopted.¹ SpicyIP presents the audit logs as a key accountability measure, although it regards them as insufficient unless similar duties are extended to the higher courts.³ Second, incident reporting: errors noticed in an approved tool's output must be reported to the Principal District Court, which forwards them to the High Court's IT department so that appropriate safety reviews can be undertaken (4.10).¹ Third, training: judges and employees "must participate" in programmes run by the Judicial Academy or the High Court on the ethical, legal, technical and practical aspects of AI (4.9).¹²

Clause 5 provides that violations "may result in disciplinary action" and that the rules on disciplinary proceedings "shall prevail", and clause 6 allows the High Court to revise the policy as technology and law develop.¹ Neither SpicyIP's analysis nor The Leaflet's explainer discussed penalties, so the express disciplinary clause in the policy text is easy to overlook.³⁴

The Kerala policy was reported as the first written, court-issued AI rulebook for Indian judges, and the sources reviewed identified no earlier one. It did three things that later instruments copied or reacted to. It put responsibility on the individual judge, so that no officer could blame a tool for an error in an order. It set a confidentiality rule for litigant data, ahead of the main obligations of the Digital Personal Data Protection Act, 2023 coming into force. And it linked tool approval to a gatekeeper, the High Court or Supreme Court, rather than leaving the choice to individual judges.¹⁵

It also sat alongside Kerala's own deployment of approved technology. Following a pilot in four Ernakulam trial courts from 1 February 2025, a memorandum of 27 September 2025 made the Adalat.AI voice-to-text tool the primary means of recording witness depositions in all Kerala courts from 1 November 2025.⁷ On 6 January 2026 the High Court announced its paperless system for Kalpetta in Wayanad, which Bar & Bench described as India's first fully paperless judicial district, with in-house AI features such as structured case summaries from the digital record, question-and-answer queries over case files and voice-to-text transcription.⁶ Taken together, the policy marks what is forbidden while the High Court builds tools it can approve.

In December 2025 Justice M.A. Abdul Hakhim observed that "some of the Writ Petitions are AI-generated and do not contain the basic material facts" in bank-account freezing cases, and that advocates were unable to explain their pleadings when questioned, a reminder that the risks of unverified AI reach the Bar as well as the Bench.⁸

Commentary has welcomed the policy while identifying gaps. SpicyIP argued that it creates an accountability asymmetry: it binds trial judges and interns but is silent on the High Court itself, and the same standards should apply "tier-agnostic" to all courts. It also warned that human-in-the-loop review can decay into ratification if judges come to rely on tools, citing research on cognitive offloading, and called for national standards on disclosure, audits and independent review.³ An SCC Online article in April 2026 made a similar point that the policy applies only to the district judiciary and urged other High Courts to adopt comparable rules.⁵

The Leaflet's review treated the policy as a rigorous codification of AI as an assistive tool, but questioned how far courts can go without digital infrastructure, judicial education and focused support from the Supreme Court and the Bar Council of India.⁴ A practical question is the supply of "Approved AI Tools". The policy depends on approval by the High Court or the Supreme Court, but the sources reviewed for this entry did not identify a published list of approved tools.

As of 15 October 2026 the policy remains the operative instrument for Kerala's district courts. The sources reviewed did not show that it has been amended or withdrawn. Two later developments may affect it.

First, other High Courts moved in April 2026. Gujarat released a Policy on Use of Artificial Intelligence in Judicial and Court Administration that bars AI from adjudication and attaches personal liability to signatories,¹⁰ and the Punjab and Haryana High Court told district judges in Punjab, Haryana and Chandigarh not to use tools such as ChatGPT, Gemini and Copilot for writing judgments or legal research.¹¹

Second, on 3 June 2026 the Supreme Court's AI Committee published draft Regulations for Use of Artificial Intelligence in Courts, 2026 for comment. These would apply to the Supreme Court, High Courts, subordinate courts and tribunals, and would come into force in each High Court on a date notified by its Chief Justice. The Kerala High Court's website lists, among its Practice Directions, a notice dated 10 June 2026 inviting comments and suggestions on the draft "Regulations for Use of AI in Courts 2026". The notice's text was not available for this entry, so how Kerala's own policy would be reconciled with the final national text is not yet known.⁹