{"id":381152,"date":"2026-04-16T10:00:35","date_gmt":"2026-04-16T04:30:35","guid":{"rendered":"https:\/\/www.scconline.com\/blog\/?p=381152"},"modified":"2026-04-16T11:49:53","modified_gmt":"2026-04-16T06:19:53","slug":"artificial-intelligence-trial-courts-india-analysis","status":"publish","type":"post","link":"https:\/\/www.scconline.com\/blog\/post\/2026\/04\/16\/artificial-intelligence-trial-courts-india-analysis\/","title":{"rendered":"Artificial Intelligence in Trial Courts: Promise, Peril, and Prudence"},"content":{"rendered":"
\n

In India, Phase III of E-Courts Project has included a component of AI to enhance case management and legal research. It has a component for future technological advancement, including AI and blockchain, across the High Courts in India up to 2027 with separate budget outlay.<\/p>\n

Judicial work is, and always has been, synonymous with application of mind. A Judge, as his core function, marshals the facts and appreciates the evidence presented before him. After applying law on such facts and evidence, a judgment is written with reasons supporting the conclusion. Reasons are the heart and soul of a judicial order. Lack of such reasons makes a decision arbitrary and betrays the most important essential of judicial determination. The advent of artificial intelligence (AI), despite its several positives, threatens the application of human mind in passing such reasoned judgments.1<\/sup><\/a> Judiciaries, across the world, are arriving at an inflection point where judges resort to AI chatbots for their research and drafting.2<\/sup><\/a><\/span> It is only a short step from there to using AI to generate judgments.<\/span><\/p>\n

The Chief Justice of India had recently assured that AI will not be used in judicial decision-making and held that “independence and human discretion of judges remain central to the Indian justice system”.3<\/sup><\/a> However, the prospect of AI being used by judges looms. Judges may resort to it to reduce their excessive burden and give in to the convenience that it provides. When presented with facts and evidence, AI systems generate a reasoned judgment with a discussion of all such facts and evidence.4<\/sup><\/a> They create efficiency and help oil the judicial machinery to enable it to dispense faster justice. But the ostensible efficiency creates several problems.<\/p>\n

In this article, we discuss the threats that emanate from the use of AI and its slippery slope which may lead the judges to use AI systems for decision-making. While it is true that AI creates convenience and efficiency in judicial functions, its use has to be guided with human control and accountability. It may appear that this is an argument against convenience, but using AI cautiously will ensure that such convenience does not undermine reasoning and propriety of a judicial decision. When such use is unguided, there is a threat of orders and judgments being passed without sufficient supervision to check the veracity of the legal positions contained in them.<\/p>\n

The architecture of AI and its expanding judicial footprint<\/h2>\n

AI systems work by analysing large amounts of data to identify patterns and relationships.5<\/sup><\/a> Based on these patterns, they produce responses or predictions to new inputs, rather than reasoning in a human and value-based manner.6<\/sup><\/a> As a result, their outputs are shaped by the data on which they are trained and the limits built into their design. Their response, however, is not parallel to human reasoning.7<\/sup><\/a> They try to mimic human reasoning based on all the information they have as to how earlier decisions have been taken.<\/p>\n

While, there are general AI systems such as ChatGPT, Perplexity, Gemini, and Grok, there have been specific systems developed for the legal field. Lucio AI is used by lawyers across the world for their legal research and drafting assistance. Harvey AI is specifically targeted at law firms with transactional practice. Spellbook helps draft contracts and Kira Systems does contract review and due diligence for mergers and acquisitions.<\/p>\n

Even judiciaries across the world have started developing AI systems intended to be used by their judges. Germany has developed OLGA to categorise cases and extract relevant case data.8<\/sup><\/a><\/span><\/span><\/span><\/span><\/span><\/span> Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) is a system used in some parts of the United States for sentencing guidance.9<\/sup><\/a> In British Columbia, Canada, a system called the “Solution Explorer” guides the litigants in the pre-trial phase by estimating their chances of success if they approach the courts.10<\/sup><\/a><\/span><\/span><\/span><\/span><\/span><\/span> China has built a similar system called “Xiao Fa”.11<\/sup><\/a> The most ambitious of them all is Victor, a system being developed by Brazil. Victor is built on a convolutional neural network (CNN). A CNN is a type of artificial neural network commonly used to analyse images and, in some cases, to process and understand language.12<\/sup><\/a><\/span><\/span><\/span> According to its developers, Victor does not make decisions on its own but merely assists court clerks in sorting and screening appeals.13<\/sup><\/a><\/span><\/span><\/span> However, there is scepticism around this as it is argued that it is difficult to draw a clear line between human decision-making and the subtle influence that software tools can exert on human reasoning.14<\/sup><\/a><\/span><\/p>\n

In India, Phase III of E-Courts Project has included a component of AI to enhance case management and legal research. It has a component for future technological advancement, including AI and blockchain, across the High Courts in India up to 2027 with separate budget outlay.15<\/sup><\/a> It includes intelligent scheduling of cases, case prediction to enable the parties predict probable outcomes of their case, natural language processing for legal research and translation of documents, and data security measures.16<\/sup><\/a> This complements the already existing tools such as Adalat AI which transcribes court proceedings such as witness depositions and oral arguments.17<\/sup><\/a><\/span><\/span><\/span> Further, an AI-based software tool called Legal Research Analysis Assistant (LegRAA) has been developed to aid judges in legal research and document analysis, but is yet to be adopted by judges in their routine work.18<\/sup><\/a> Another tool called Supreme Court Portal for Assistance in Court’s Efficiency (SUPACE) has been created to aid judges by retrieving facts from documents submitted by the litigants and providing relevant case laws.19<\/sup><\/a> However, its adoption has been limited as it is only used for criminal matters before the Delhi and Bombay High Courts.<\/span><\/p>\n

While these initiatives are undoubtedly welcome, they require careful and cautious engagement. There is little dispute that AI can enhance efficiency and accelerate the disposal of cases in trial courts. Yet, this very promise compels a deeper inquiry into its limits. To what extent should AI be permitted to operate within the judicial process? What happens when systems designed for assistance begin to be used beyond their intended scope? And how does the legitimacy of adjudication fare if judges increasingly rely on AI for decision-making rather than independent judicial reasoning? Most importantly, why is human reasoning better than reasoning supplied by AI in drafting orders and judgments? These questions must be answered before AI is routinely used by the trial courts judges.<\/p>\n

Limitations and risks of AI in judicial decision-making<\/h2>\n

Adjudication in trial courts is fact-heavy and fact-specific. Merely because two cases have similar facts, the result cannot be the same. A large language model learns from the vast data available on the internet and responds by identifying similarities.20<\/sup><\/a> This can create a problem of generalisation even though the specific facts at hand deserve an independent deliberation.21<\/sup><\/a> For example, when it encounters an interlocutory application for temporary injunction that appears similar to another application found on the internet, it may treat both of them as identical and suggest the same outcome. Hence, if the petition found on the internet was allowed, the model is likely to recommend allowance of the subsequent one as well. However, pleadings may be superficially similar while the underlying facts and surrounding circumstances differ materially. Judicial determination requires each factual situation to be assessed independently, without mechanical reliance on precedent or pattern. The use of AI in such contexts therefore risks undermining such application of independent judicial mind, which is a core foundation of adjudication.<\/p>\n

To ensure that each case is decided on its own merits, the documentary evidence produced during the trial of such case has to be perused. A significant concern arises when AI systems are expected to extract and evaluate facts from such documents, a task that lies at the very core of a trial Judge’s function. Judges are required to peruse a wide and often unstructured range of documents, ranging from panchanamas to land records maintained in regional languages, such as Pahanis or Adangals in Telugu-speaking States. In many instances, these documents are handwritten, poorly preserved, or contain overwriting, corrections, and interpolations. An AI system may struggle not only to accurately extract data from such records, but also to comprehend their legal significance. Optical character recognition (OCR) using AI is not foolproof and is susceptible to errors arising from document quality, formatting, and language complexity.22<\/sup><\/a> Even where extraction is attempted, a deeper question arises as to whether the system can correctly contextualise those facts in relation to the issue under consideration. Corrections in documents, or explanations offered during oral evidence, such as an investigating officer clarifying the absence of date in an inquest panchanama, require careful human evaluation. Further, a substantial portion of judicial records in India are not machine-readable.23<\/sup><\/a> The assessment of their probative value necessarily demands independent application of the human mind, which cannot be delegated to AI systems.24<\/sup><\/a><\/p>\n

When an order or a judgment is generated using AI system, the Judge is not articulating his reasoning behind the decision. Rather, he is merely confirming the reasoning stated by the AI system. This converts the job of a judge from “application of mind” to produce a well-reasoned result to “confirmation of output” of the AI. Hence, instead of constructing the reason, a judge is simply auditing the reason. If such reason is well-articulated, a judge may find it compelling to approve it, rather than take the efforts to finding any flaws to correct such reasoning before pronouncing the judgment.25<\/sup><\/a> When this becomes routine, the intensity of how minutely the reason is reviewed gradually diminishes. It betrays the essential feature of a judicial function which is to decide by the application of human mind.26<\/sup><\/a> This will lead to a creation of “black box” where one is unable to ascertain why an AI system has produced an order in a particular manner.27<\/sup><\/a> Some have argued that a judges should be provided with a dedicated AI tool, called “Explainable AI” which also explains the reasons behind a determination made by it, and until such time, judges should be restricted to delegate their reasoning functions to an AI system.28<\/sup><\/a><\/p>\n

As the output of AI barely takes any time, a judge may not dedicate several hours needed to peruse through the evidence with an open mind and find out if the losing side has any strength in its case. It is often seen that a doubt created during dictation of a judgment leads a judge to ponder on whether such judgment is going on the right track. However, AI does not provide any such opportunity as it cuts shorts the efforts of the judge to take a deep dive into the evidence. It is for this reason that the House of Lords stated in its Select Committee Report that “it is not acceptable to deploy any artificial intelligence system which could have a substantial impact on an individual’s life, unless it can generate a full and satisfactory explanation for the decisions it will take”.29<\/sup><\/a><\/p>\n

When it becomes public knowledge that a judge uses AI systems to produce his orders or judgments, it casts a blow to public trust in judiciary. It erodes trust in the judicial system where human reasoning is expected to be applied. This undermines the prestige of the judiciary as it will seem that the core judicial function has been outsourced to a computer software.30<\/sup><\/a> This also provides a new ground for appeal where a new wave of appeals may be preferred on the ground that the judgment of the trial court was not written by the Judge himself, but by an AI system.<\/p>\n

A fundamental ethical concern arises as the process risks diluting accountability and the moral responsibility that lies at the heart of adjudication. The European Union Artificial Intelligence Act, 2024<\/span><\/span><\/span>, adopts a risk-based regulatory framework that classifies AI activities into four categories: banned, high-risk, limited risk, and minimal risk.31<\/sup><\/a> Notably, the deployment of AI for legal research, interpretative assistance, and the application of law to facts is designated as “high-risk”. Reflecting the sensitivity of adjudicatory functions, the Act mandates meaningful human oversight wherever AI is used within judicial processes, underscoring that technology may assist but cannot supplant human judgment. There is no such requirement in India, and even when such requirement is enacted as a law or mandated by rules of practice, ensuring that there is adequate human oversight to not undermine application of mind by the judges is difficult.<\/span><\/p>\n

The Delhi High Court sounded this alarm when it held that “there is no doubt in the mind of the court that, at the present stage of technological development, AI cannot substitute either human intelligence or the humane element in the adjudicatory process. At best, the tool could be utilised for preliminary understanding or for preliminary research, and nothing more<\/span><\/span><\/span>.”32<\/sup><\/a><\/span><\/span><\/p>\n

Another serious ethical concern arises when trial courts rely on AI-generated legal research, as such systems may fabricate authorities and reinforce the user’s pre-existing belief rather than accurately stating the law. At times, AI software “hallucinates” case law by creating fictitious judgments that appear plausible and authoritative and presents them confidently to affirm what the user seeks.33<\/sup><\/a><\/span><\/span><\/span> Upon verification on authentic legal databases such as SCC, these cases are found not to exist at all. Such “ghost” case law grossly misrepresents legal principles, and if relied upon by trial courts, may result in orders based on non-existent decisions of the Supreme Court or High Courts. To illustrate this, consider the following prompt that was inserted by the authors on ChatGPT:<\/span><\/p>\n\n\n
\n

Prompt: Give me a case law which guides appointment of receivers and for what purpose a receiver can be appointed?<\/i><\/p>\n

ChatGPT Reply:<\/p>\n

Certainly! One of the leading judgments guiding the appointment of receivers under Order XL Rule 1 CPC<\/b> is:<\/p>\n

\n

\ud83d\udc68\u200d\u2696\ufe0f Krishna Kumar Birla v. Rajendra Singh Lodha,<\/b><\/p>\n

(2008) 4 SCC 300<\/a><\/b><\/p>\n

\ud83d\udccb Background:<\/b><\/p>\n

This case involved a dispute over the vast estate of late industrialist Priyamvada Birla. There was a will, and parties were fighting over its validity. One side sought appointment of a receiver to protect the estate from being misused during litigation.<\/p>\n

\u2696\ufe0f Key Principles Laid Down:<\/b><\/p>\n

The Supreme Court held that:<\/p>\n

“Appointment of a receiver is a drastic and harsh remedy and should be exercised with extreme caution and only when there is a strong prima facie case and a clear danger of loss or damage to the property in dispute.<\/b>“<\/i><\/p>\n

The Court summarized that:<\/p>\n

    \n
  1. Appointment should be just and convenient<\/b> — it must be in the interest of justice.<\/li>\n
  2. A receiver is not appointed merely because there is a dispute between the parties.<\/li>\n
  3. It must be shown that:\n