The Gujarat High Court has announced a policy governing the use of artificial intelligence (AI) in judicial administration and has taken note of the following points: need to be regulated this Technology is rapidly expanding within the justice system. The court pointed out that AI increases efficiency Research and management also involve risks such as: Overdependence, bias, and erosion of human judgment.
Because the judiciary’s constitutional mandate requires it to use human reasoning and accountability to administer justice, the court emphasized that the judiciary must establish standards such as: strict framework To prevent AI violation of judgment. Violations may result in disciplinary action and civil or criminal liability.
Scope and scope:
- applies to all Judicial officers, court staff, interns, legal assistants, and related agencies.
- covers both High Court and District Court function.
- extend to Public hearings, case management, registration services, investigations, and administrative services.
- Applies whether or not work is done On court premises or at a remote location.
- Covers use in court systems, personal devices, third-party platforms, cloud tools, mobile applications, and browser extensions.
- What’s included All AI tools: (Large-Scale Language Models) AI capabilities built into productivity software such as LLMs (ChatGPT, Gemini, etc.), legal research tools, translation systems, transcription tools, summarization tools, predictive analytics, or word processors.
Permitted AI uses:
- Administrative tasks IT automation, training materials, drafting circulars and notices based on public domain information, etc. are aimed at improving internal productivity.
- legal investigation support; This includes searching and analyzing judgments, identifying precedent, and interpreting laws, provided that the output is verified against authoritative sources.
- Drafting aid It is limited to improving the language, structure, and clarity of orders and judgments and must not affect substantive reasoning or legal conclusions.
- Generating a structural outline or framework Decisions and opinions are subject to full judicial review and complete rewriting by judges.
- translation and transcription Creating documents and hearings using AI tools; Required validation and authentication By competent human authority before use.
- Case management featuresScheduling, fair case assignment, management reporting, and more are based solely on anonymized metadata such as case type, workload, and abandonment rate.
Prohibited uses of AI:
- Any use of AI judicial decision making, reasoning, rulings, or application of law; Includes bail, sentencing, interim orders, and final judgments.
- drafting, producing, or substantively composing a judgment;an order, or a binding legal judgment, even if it is later reviewed by a judge.
- determining facts, interpreting or weighing evidence; Assessing credibility or summarizing depositions and testimony in judicial proceedings.
- Input confidential information into publicly available AI tools, This includes the identity of the parties, details of pending cases, privileged communications, or sensitive personal data.
- Creation, fabrication, alteration, and embellishment of evidence In any form using AI systems.
- Using AI-generated citationslegal provisions without case law references or independent verification from authoritative primary sources.
- Drafting, revising, and summarizing internal memos or submission using AI tools.
- Access AI tools by: Bypassing court security protocols.
Confidentiality and data protection: This policy prohibits users from: Input sensitive or incident-related data into public AI tools. This limits the publishing tool to general tasks that are not case-specific. parable Approved enterprise AI systemthe user must: Don’t enter sensitive data Witness identities, confidential court information, special categories of personal data, etc. All AI-related processing is Comply with the Digital Personal Data Protection Act 2023Users must not reuse personal data beyond the original judicial purpose.
Verification and accountability: policy Mandating rigorous verification Analyze all output generated by AI, including citations to primary sources, legal proposals, summaries, translations, and transcripts. The user must: independent of format and validity It is used as proof of accuracy and the original document should be reviewed before acting.
judges and court staff take full personal responsibility For all outputs including AI-assisted work. They must disclose the use of AI in studies and preparatory documents and cannot blame errors on AI. Further AI utilization not alleviate Liability for Inaccuracies, Misconduct, or Negligence.
Other policies of the High Court: The Punjab and Haryana High Court has prohibited judicial officers in Punjab, Haryana and Chandigarh from using AI tools such as ChatGPT and Gemini for writing judgments and conducting legal research. The directive issued to District Judges and Sessions Judges warns that any violation will be treated seriously by the High Court authorities. Similarly, in July 2025, the Kerala High Court prohibited district court judges from using AI tools to pass any findings, remedies, orders, or judgments.
The policy does not allow for data bias. This policy does not say how Underlying data quality can Forming administrative results using AI. The February 2026 DFL report highlights: judicial dataset often incomplete, inconsistent, not representativeincreasing the risk of output distortion.
Even if the tools focus on efficiency, such as: Scheduling or prioritizing casescan produce discriminatory results If it is built on limited or uneven metadata, it can disadvantage certain litigants or litigation categories.
Despite allowing case allocation based on metadata, Gujarat’s policy does not mandate auditing, bias testing, or representativeness of datasets check. That too Skips system-level protection measures These include audit logs, explainability requirements, and the organizational technical oversight required to detect and remediate such risks.
Please also read
