NextFin News - The Punjab and Haryana High Court has issued a sweeping prohibition against the use of artificial intelligence tools by judicial officers, marking a significant escalation in the legal system’s resistance to automated adjudication. In a directive issued on April 6, 2026, the court’s registrar-general informed district and sessions judges across Punjab, Haryana, and Chandigarh that tools including ChatGPT, Google’s Gemini, Microsoft’s Copilot, and Meta AI are now strictly off-limits for both judgment writing and legal research. The order, issued at the behest of the Chief Justice, carries a stern warning that any violation will be "viewed seriously," signaling a zero-tolerance policy toward the encroachment of large language models into the core of the judicial process.
This administrative crackdown follows a similar move by the Gujarat High Court just days earlier, which unveiled a comprehensive AI policy on April 5. While the Gujarat policy allows for AI to be used to enhance administrative speed, it explicitly bars the technology from "substantive adjudicatory processes," including bail sentencing and judicial reasoning. The Kerala High Court had set a precedent for such restrictions in July 2025, but the simultaneous actions by two major northern and western Indian high courts this week suggest a coordinated institutional pushback against the rapid, unregulated adoption of generative AI in the lower courts.
The primary concern cited by judicial authorities is the preservation of human reasoning, a cornerstone of the common law tradition that AI currently lacks. According to the Gujarat High Court’s policy, the risks of "over-reliance, bias, and the erosion of human judgment" outweigh the efficiency gains offered by automation. Legal experts note that while AI can summarize vast amounts of case law in seconds, it is prone to "hallucinations"—the generation of plausible-sounding but entirely fabricated legal citations—which could lead to miscarriages of justice if not meticulously verified by human eyes.
Data security and confidentiality form the second pillar of the ban. The Punjab and Haryana directive effectively treats public AI tools as a threat to the sanctity of the judicial record. By inputting case details or draft judgments into these platforms, judicial officers risk leaking sensitive information into the training sets of private corporations. The Gujarat policy specifically prohibits entering any confidential or case-related data into public AI tools, restricting their use to general, non-specific administrative tasks. This reflects a growing global anxiety among institutional leaders regarding the "black box" nature of proprietary AI models and the lack of transparency in how they process and store user data.
However, the ban is not without its critics within the legal tech community. Some practitioners argue that a total prohibition on AI for legal research may be counterproductive, given the sheer volume of pending cases in the Indian judiciary—which currently exceeds 50 million across all levels. Proponents of "Legal AI" suggest that instead of a blanket ban, the judiciary should develop secure, closed-loop AI systems trained specifically on verified Indian statutes and precedents. They argue that properly calibrated tools could help clear the massive backlog by automating routine drafting tasks, leaving judges more time to focus on complex legal interpretation.
The current stance of the High Courts suggests that the judiciary is prioritizing institutional integrity over technological speed. By drawing a hard line at the drafting of judgments, the courts are asserting that the "human element" of justice—empathy, moral nuance, and accountability—cannot be outsourced to an algorithm. As the legal landscape continues to grapple with these tools, the focus is likely to shift toward the development of internal, sovereign AI frameworks that can offer the benefits of automation without compromising the privacy or the intellectual independence of the bench.
Explore more exclusive insights at nextfin.ai.
