As John Rawls was penning down that, “A legal system must make provisions for conducting orderly trials and hearings; it must contain rules of evidence that guarantee rational procedures of inquiry. While there are variations in these procedures, the rule of law requires some form of due process: that is, a process reasonably designed to ascertain the truth, in ways consistent with the other ends of the legal system, as to whether a violation has taken place and under what circumstances.”[i], he could hardly have imagined that the tenets of “Rational Procedures of Inquiry” and “Due Process reasonably designed to ascertain the truth” would be qualified and weighed by a judge who is not a human. The assumption would’ve been a far-fetched or implausible possibility, as it seems though, since the Law is meant to govern humans by a society comprised of humans themselves.

As we read this article, we stand at an inflection point where Artificial Intelligence stands at the doorstep of the custodians our laws and human rights. AI is still in its nascent stage to be introduced in the legal arena and infamous for producing notorious outputs, which don’t even exist. The Supreme Court has already raised serious concerns regarding these AI “hallucinations”. The problem gets aggravated by the rampant use of AI by litigants and advocates, which can hamper the delivery of justice.

The events take an alarming turn when we come across the impetuous use of AI tools by the bench and seeing the justice getting hijacked. In March this year, the Supreme Court stayed a lower court’s order after discovering that it was adjudicated using fake judgements generated by artificial intelligence.

In the wake of these developments, the Artificial Intelligence Committee of the Supreme Court released its Draft Regulations for Use of Artificial Intelligence in Courts, 2026 on 3 June 2026 in an attempt to lay down the guiding principles for regulating the use of AI tools in court proceedings. The guidelines allow for the use of AI tools in the court in an “assistive capacity”.

Beyond efficiency: a Constitutional Question

The most compelling argument in support of introducing AI in courts is based on its efficiency of conducting legal research, scheduling cases, summarising records, translating documents, transcribing proceedings and organising case management with remarkable speed. With the courts loaded with excessive workload and pendency of cases across the hierarchy, AI seems to offer a promising potential to hasten the process of clearing backlog.

However, the notion is not free from hazards. A typical AI software builds its outputs upon the data collected through the resources stored in its database called the Large Language Modules. Its reasoning is built on the patterns identified in the data rather than a profound normative reasoning like humans. Thus, it can only mimic human reasoning by observing the earlier patterns, without judging the aptness of the output in the case at hand. It is due to these issues that nations across the globe are taking utmost precautions before allowing the use of AI tools in their legal systems.

An AI system cannot be completely impartial in the constitutional sense because its outputs are a function of the data on which it is trained, which reflects the social and institutional biases of the societies that produced it. It cannot be said to be independent because it carries, invisibly and indelibly, the design choices and commercial decisions of its developer.

Article 21 of the Constitution guarantees that no person shall be deprived of life or personal liberty except by a procedure that is fair, just, and reasonable. It is difficult to see how an AI system, working on algorithms which may be fraught with historical biases or wrongful decisions of the earlier courts, which cannot even explain the reasoning of its output or check whether the judgments relied upon are existent or not, let alone checking whether they are overruled, can live up to the cardinal principles laid down in Article 21 of the Constitution, and upheld by different landmark judgments.[ii]

Further, in TMA Pai Foundation v. State of Karnataka[iii]  as well as Ramana Dayaram Shetty v Industrial Airport Authority of India,[iv] it has been professed that the any arbitrary procedure that strips off the very crux of justice delivery process and cannot trace or justify the reasoning of the decision, cannot sustain the tests of fair procedure and reasonableness enshrined under Article 14. AI systems don’t have the capability to act upon the ever-evolving principles of natural justice and procedural fairness that can adapt to new social realities. 

Introducing AI in the courts is fraught with risks of algorithmic bias. It tends to ignore the central facets of justice delivery system. Efficiency alone has never been the constitutional objective of the judicial process. Courts are not factories designed to maximise output. Judicial legitimacy rests not merely upon the correctness of decisions but upon the fairness of the process through which those decisions are reached.

Adjudication involves something more than Data Processing

For the losing party, a judgment often appears to be the pronouncement of judges sitting in ivory towers. Yet few appreciate what goes on in these “ivory towers” before arriving at a decision.

A judge has a core duty to record, evaluate, apply and weigh all the documentary evidence in the face of the specific facts of the case. The facts need to me admissible in the eyes of law in order to gain legitimacy. According to section 63(4)(c) of the Bharatiya Sakshya Adhiniyam, 2023, a certificate from an authorized officer has to be attached for making an electronic evidence admissible. The provision is further qualified by section 79A of the Information Technology Act, 2000, which mandates the Central government to appoint an authority which can offer expert analysis of electronic evidence before they are presented in a court of law.[v] The legal challenge faced by the authorities is how an AI system will be given authority to access and admit these documentary evidence without a legal backing. Given the sensitive, multilingual and often unstructured nature of evidence in the lower judiciary in India, AI can hardly be trusted to interpret the documents produced by the parties without any errors and in the intended context. Additionally, all the documentary evidences have to be evaluated in the light of oral pleadings of the counsel on behalf of the parties which can only be evaluated by a human mind.

Adjudication is more complicated than a simple application of laws to the facts of the case. The judge also has to take other equally important considerations into account while interpreting certain terms like “reasonable,” “proportionate,” “in the interest of justice,” and “best interests of the child” etc. which cannot be mechanically construed. They demand interpretation that uphold constitutional values, particular to the circumstances of the parties, and normative commitments that no statute can capture completely.

Thus, a structural tension that AI adjudication brings to the surface is the tension between “equitable justice” and “codified justice”. The former is contextual, discretionary, and attentive to the individual, while the latter values standardisation and quantifiable consistency.[vi]

Human Judge as an element of a Fair Trial

The right to a fair trial has traditionally been understood through procedural guarantees. Inherently, the principles of fair trial and rule of law are closely linked with human rights protection.[vii] Depriving the accused or the prosecution of a fair hearing undermines even the minimum standard of due process of law under Article 21 of the Constitution. A fair trial must be a real one, not a sham or mere farce of pretence.[viii]

Thus, the courts can call out any practice or procedure that casts a doubt upon the sanctity of justice delivery system. Afterall, justice must not only be done, but seen to be done. Litigants do not merely seek a befitting legal answer; they seek to be heard, understood and judged by another human being who bears constitutional responsibility for the decision rendered.

Thus, it is not hard to imagine that a right to be tried by a human judge can take shape in the future as a part of fair trial and a safeguard against adjudication by AI systems.

The right to a human judge does not need to be invented. It is implicit in every constitutional guarantee of fair hearing. A litigant who comes before a court is entitled to be heard and decided upon by a human being who carries real moral responsibility for the decision. The shift from human judgment to algorithms is a shift from justice to processing. This shift threatens to touch the legitimacy of the entire judicial enterprise, and it should be treated with corresponding seriousness.^[ix]

The Constitution does not explicitly declare such a right. Nevertheless, constitutional rights often emerge through judicial interpretation. The right to legal aid, the right to privacy and numerous procedural safeguards under Article 21 were recognised not because they were expressly written into the Constitution but because they were indispensable to ensuring fairness.

Similarly, one may argue that the identity of the decision-maker itself forms part of procedural fairness, since a person has a right not to be tried by a particular judge, under right to fair trial if there are chances of a real or apparent bias against him. This can be extended to include right not to be tried by a non-human entity for the purpose of incarceration, if it is shown that an individual has been incarcerated due to him being classified under a certain class of person by an AI system, who are prone to commit an offence.

The Cost of Algorithmic Justice

The draft regulations are probably the most consequential institutional step to harmonize AI and courts by far, arriving at a moment of acute anxiety. The pace at which AI is taking root into the legal systems across the world is actually disturbing. However, the question that demands an honest, principled answer is not ‘How do we use AI in courts?’ but ‘At what cost?’

Human life is of utmost importance, which is why there are strict procedural requirements for incarceration at the trial stage alongside various safeguards at different stages of the procedure. The first line of casualty will be the sanctity of these procedural safeguards, since the decisions are now devoid of human considerations.

The authorities and courts will be flooded with appeals arising out of procedural irregularities resulting from AI hallucinations. Rule 8 of the draft regulations of the Supreme court places the accountability of AI based decisions on the judicial officers acting upon it, irrespective of the fact that the response was hallucinated or due to the opaqueness of the “black box”. This is likely to open the floodgates of new prosecutions and disciplinary proceedings against our existing judicial task force.

A greater risk posed by the unchecked use of AI hallucinated judgments lingers on the precedential value of judgments. If not repealed on time, such decisions can give a compelling value to societal narratives that tend to be biased with respect to minority communities or a particular set of individuals.

Aforementioned factors collectively undermine the public trust in the judiciary. People expect a human reasoning, backed not only with evidence but also moral considerations, to be applied to their grievances to arrive at a just and reasoned settlement[x]. The right to be adjudicated by a human mind rests at the heart of justice delivery and the cost of outsourcing this central task to a non-human entity will cause serious damage to judicial prestige.

Risks of Giving in to Ease of Automation

Although the Judiciary has adopted a cautious approach in its draft regulations, and is equally optimistic of maintaining an arm’s length with AI, the early traces of intrusion are already visible.

The least precaution we can take against is not to give into the ease of automation. When a judge uses AI to draft the reasoning in a judgment, the judicial function shifts from the active construction of reasons to the passive confirmation of an output. A well-articulated AI-generated draft creates a greater likelihood of endorsement by the judge. What begins as critical review becomes a cursory approval, over time. The judge who once spent hours wrestling with conflicting evidence now spends minutes approving a machine’s synthesis. The intensity of engagement diminishes and so does the quality of justice.

Making matters worse, the very act of reading a fluently constructed AI draft changes how a judge approaches the reasoning. The mind moves to evaluate a proposition already framed, rather than building the proposition from the ground up.

Though accountability and the primacy of judicial decision through human intellect has been abundantly stressed in the draft regulations of the Supreme Court, a more prudent approach should be adopted by introducing the AI in courts in civil matters at first, rather than implementing it lock, stock and barrel, since the road to AI criminal law courts and other courts dealing with sensitive matters like Constitutional law, is fraught with higher and irreversible hazards.

Furthermore, a draft instrument with aspirational principles is not the same as a binding legal framework with enforcement teeth. The mechanism for detecting and sanctioning a judge who silently relies on AI output without disclosure is not adequately specified and needs certain improvements before landing in courts.

[i] John Rawls, A Theory of Justice (first Published 1921, The Belknap Press of Harvard University Press, Cambridge, 1999) 210

[ii] Maneka Gandhi v. Union of India AIR 1978 SC 597, (1978) 1 SCC 248

[iii] TMA Pai Foundation v. State of Karnataka AIR 2003 SC 355, (2002) 8 SCC 481

[iv] Ramana Dayaram Shetty v International Airport Authority of India AIR 1979 SC 1628

[v] Madhumitha C, Sandhiya Krishnan S, ‘Transition from Human to AI-Judge – The Future Automated Adjudication Mechanism in India’ (2024) 12(9) IJRASET <https://www.ijraset.com/research-paper/transition-from-human-to-ai-judge-the-future-automated-adjudication-mechanism-in-india#references>

[vi] Re, Richard M. & Solow-Niederman, Alicia, ‘Developing Artificially Intelligent Justice’ (2019) STLR 242, 248-252

[vii] Himanshu Singh Sabharwal v. State of Madhya Pradesh and others (2008) 2 SCC (Cri) 106, AIR 2008 SC 1943 [5]

[viii] Zahira Habibullah Sheikh v. State of Gujarat (2006) 2 SCC (Cri) 8, (2006) 3 SCC 374 [36], [38]

[ix]Madhumitha C, Sandhiya Krishnan S, ‘Transition from Human to AI-Judge – The Future Automated Adjudication Mechanism in India’ (2024) 12(9) IJRASET <https://www.ijraset.com/research-paper/transition-from-human-to-ai-judge-the-future-automated-adjudication-mechanism-in-india#references>

[x] Benjamin Minhao Chen, Alexander Stremitzer & Kevin Tobia, ‘Having Your Day in Robot Court’ (2022) HJLT 127, 147