In July this year, the Kerala High Court published a set of guidelines regarding the use of artificial intelligence (AI) by district judiciary (“Policy on the Use of Artificial Intelligence Tools in District Judicial Judicial”). The country's first policy is timely, as it directly addresses the use of AI in the judicial process and directly addresses strict safeguards. AI tools from document translation to filing identification are expected to improve speed and efficiency.
There's a problem
However, apparently harmless tasks like AI-enabled translation and transcription are not without risk. For example, a judge of the Supreme Court of India reported the translation of “permission” to “chhutti sweekaar (approved on holidays). For Noel Anthony Clarke vs Guardian News & Media Ltd. (2025) EWHC 550 (KB), the AI transcription tool repeatedly transcribed the claimant's name, “Noel” with “No.” It has been reported that whispers from Openai, an AI-powered voice recognition system, sometimes supplement or “hastised” the entire phrase and sentence, especially when people speak at longer pauses between words.
Search engine bias in AI-enabled legal research could tweak users to results affected by user patterns and could “watch” relevant precedents. A study published in the Journal of Empirical Legal Study found that a large legal language model (LLM) can supplement case law and demonstrate claims by citing false sources.
On a more structural level, AI risks reducing adjudication on rule-based inference, overlooking a combination of precedents, specific contexts, and relevances that influence judicial decision-making.
Currently, several market tools are used in courts on a non-commercial test basis, such as oral discussions and transcription of witness deposits. Without a specified timeframe, success parameters, or a framework for private, sensitive, or access, storage and use of personal data, such pilots should be carefully considered. AI tools were provided to courts based on risks of creating dependencies without a clear pathway to sustainable adoption. Furthermore, new technology paradigms require critical infrastructure such as trusted internet connectivity and hardware.
A rapid analysis of publicly available bids for AI services across courts shows that even if adoption is prudent, the courts do not necessarily design risk management frameworks to address ethical and legal risks. Human checks and balances such as manual review of AI translation decisions by retired judges have been introduced by advocates and translators, but AI systems can learn from available data and encounter new information in new contexts. Scholars should note that hallucinations of LLMS are functional and require careful adoption in human surveillance and high-risk scenarios, rather than bugs.
As courts increasingly integrate the use of AI in their daily operations, the combination of AI's ethical risks and legal system complexity requires effective guardrails to mitigate risk. As the majority of court proceedings remain paper-based, the transition to advance AI deployment should not further debilitate an already incomplete system.
First, there is a need for important AI literacy among judges, court staff and lawyers. In addition to building the capacity to use AI tools, programs are also required to understand the limitations of deployed systems. Working with AI governance experts, the Judicial Academy and Bar Association are well placed to promote such capacity building.
Second, guidelines are needed to shape the individual use of generated AI for research and judgment writing. If AI is used in the arbitration process, you must have the right to notify the litigator. Similarly, litigators and lawyers have the right to know whether AI is being used in certain courts. Given the possibility of errors arising from the use of AI, courts should look into whether pilot opt-out or full deployment AI litigators are permitted if they have concerns about protection or human surveillance.
Third, courts should adopt standardized procurement guidelines to support assessment of the reliability and suitability of proposed AI systems for tasks at hand. Pre-cooking measures can also help the court diagnose the exact problem and whether AI is the best solution. The procurement framework can guide the assessment of technical standards for explanability, data management, and risk mitigation.
About the Ecourts Project
These frameworks allow decision makers to monitor vendor compliance and performance. This may go beyond the routine expertise of judges and registry.
The Phase III vision document of the Ecourts Project (e-Committee, Supreme Court of India) recognizes the need to create a technology office to evaluate, select and oversee the implementation of complex digital solutions such as infrastructure and software. Such a scaffolding to assist and support decision-making regarding the use and recruitment of AI is one way to overcome gaps in technical expertise. A dedicated expert can give courts clearer guidance when adopting AI tools as part of a comprehensive plan.
As courts start to move towards adopting AI, it is important not to lose sight of the ultimate purpose of AI within the system. In this rapidly evolving technological landscape, clear guidelines for the use and adoption of AI in courts are essential to ensure that the promotion of an efficient court system does not overturn the subtle inference and human decision-making that is at the heart of the arbitration process.
Ria Verghese works in Daksh, Bengaluru. Smita Mutt works for Daksh, Bengaluru. Dona Mathew works at Digital Futures Lab in Goa
Published – August 23, 2025 12:08 AM IST
