Artificial Intelligence and Judicial Decision-Making: Legal, Ethical and Constitutional Perspectives

ABSTRACT

Artificial Intelligence (AI) is migrating from the periphery of legal administration to the operational core of judicial systems worldwide, assisting or, in some jurisdictions, materially influencing decisions on bail, sentencing, and case triage. This paper undertakes a doctrinal and comparative examination of the legal, ethical, and constitutional implications of this migration, with particular reference to the Indian constitutional order. Drawing on the due-process critique developed in the American case of State v. Loomis, the risk-based classification of judicial AI as “high-risk” under the European Union’s Artificial Intelligence Act (Regulation (EU) 2024/1689), and India’s own institutional experiments — SUPACE, SUVAS, LegRAA, and TERES, developed under the e-Courts Mission Mode Project — the paper argues that unregulated or under-regulated AI deployment in adjudication poses a substantive risk to the guarantees of equality (Article 14) and due process (Article 21) under the Constitution of India. The paper critically evaluates the Supreme Court of India’s Draft Regulations for the Use of Artificial Intelligence in Courts, 2026, and proposes a graduated, four-tier regulatory framework — assistive, research-support, decision-support, and prohibited — calibrated to Indian constitutional values and informed by comparative best practice. The paper concludes that AI’s legitimate role in Indian adjudication lies in relieving administrative and research burdens, never in displacing the constitutionally mandated exercise of independent human judicial judgment.

Keywords

Artificial Intelligence · Judicial Decision-Making · Due Process · Article 21 · Algorithmic Bias · SUPACE · EU AI Act · State v. Loomis · Judicial Independence · Constitutional Law

INTRODUCTION

The judicial function has traditionally been understood as an exercise of human reason, conscience, and discretion exercised within the framework of law and precedent. Article 14 of the Constitution of India guarantees equality before law, and Article 21 guarantees the right to life and personal liberty, which the Supreme Court has repeatedly held to encompass the right to a fair trial and access to justice[1]. The legitimacy of adjudication has, until recently, rested on the assumption that a human judge — bound by oath, subject to appeal, and capable of articulating reasons — stands at the centre of the decisional process. The emergence of Artificial Intelligence (AI) as a tool of judicial administration disturbs this settled assumption and compels a re-examination of foundational doctrines of due process, natural justice, and the rule of law.

AI, understood broadly as computational systems capable of performing tasks that would ordinarily require human intelligence — including pattern recognition, natural language processing, and predictive inference — has moved from the periphery of legal administration to its operational core in several jurisdictions. In the United States, actuarial risk-assessment tools such as COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) have been used at the sentencing and bail stage since the early 2010s; their constitutionality was tested in the celebrated case of State v. Loomis[2]. In the United Kingdom, the Durham Constabulary’s Harm Assessment Risk Tool (HART) has assisted custody decisions. China has invested in “Smart Court” and “Xiao Zhi” systems that assist judges with case-law retrieval and sentencing-consistency checks. Estonia has explored a limited “robot judge” concept for small-value contractual disputes. At the supranational level, the European Union’s Artificial Intelligence Act (Regulation (EU) 2024/1689), in force since 1 August 2024, classifies AI systems intended to assist judicial authorities in researching and interpreting facts and law, and in applying law to a concrete set of facts, as “high-risk”[3] — attracting stringent transparency, human-oversight, and accountability obligations.

India’s engagement with judicial AI has proceeded along a deliberately compartmentalised and assistive trajectory under the e-Courts Mission Mode Project. SUVAS (Supreme Court Vidhik Anuvaad Software), launched on Constitution Day, 26 November 2019, is a neural machine translation system that has translated more than 36,000 judgments into regional languages[4]. SUPACE (Supreme Court Portal for Assistance in Court Efficiency) assists judges by organising the factual matrix of a case and surfacing relevant precedents, without generating recommendations[5]. LegRAA (Legal Research Analysis Assistant) and TERES (real-time transcription) extend this assistive architecture[6]. The Supreme Court of India’s Draft Regulations for the Use of Artificial Intelligence in Courts, 2026, published on 3 June 2026, mark India’s first binding, national, cross-tier attempt to govern this emerging landscape[7].

This paper proceeds in six further parts. Part 2 reviews the relevant literature. Part 3 outlines the research methodology. Part 4 undertakes a comparative survey of judicial AI across major jurisdictions. Part 5 examines India’s institutional experience in detail. Part 6 subjects this comparative and institutional material to constitutional and ethical analysis. Part 7 presents findings and a proposed regulatory framework, followed by a concluding part offering recommendations.

Objectives of the Study

• To examine the constitutional foundations of judicial decision-making in India and their adequacy in regulating AI-assisted adjudication.
• To analyse comparative regulatory approaches to judicial AI in the United States, the European Union, the United Kingdom, and China.
• To evaluate India’s judicial AI infrastructure against constitutional and ethical benchmarks of transparency, non-discrimination, and accountability.
• To critically assess India’s Draft Regulations for the Use of Artificial Intelligence in Courts, 2026.
• To propose a doctrinally grounded, graduated regulatory framework for the constitutionally compliant integration of AI into Indian judicial administration.

REVIEW OF LITERATURE

The literature on AI and judicial decision-making has developed rapidly since the mid-2010s and can be organised into five broad, overlapping strands.

2.1 Foundational Techno-Legal Scholarship

Early scholarship on “algorithmic justice” focused on the descriptive question of where and how predictive algorithms were entering criminal justice systems. Commentary emerging from the Loomis litigation — including analysis in the Harvard Law Review and the Harvard Journal of Law & Technology — established the now-standard critique that risk-assessment tools of the COMPAS variety raise due-process concerns because their proprietary architecture forecloses meaningful adversarial testing, and because actuarial accuracy rates (assessed by some commentators at roughly seventy percent) sit uneasily with a criminal process that prizes individualised, non-statistical justice[8].

2.2 Algorithmic Bias and Discrimination Studies

A second strand, exemplified by ProPublica’s investigative analysis of COMPAS, demonstrates that facially neutral predictive tools can encode and reproduce historical patterns of discriminatory enforcement and sentencing, particularly along racial lines in the American context[9]. This literature is methodologically significant because it demonstrates a replicable empirical technique — comparing false-positive and false-negative rates across demographic sub-groups — with clear relevance to any future audit of Indian judicial AI tools.

2.3 Comparative Regulatory Scholarship

The European Union’s AI Act has generated a substantial body of legal-analytical literature, including sectoral treatment in the Cambridge Handbook of AI and Technologies in Courts, examining the classification of judicial AI systems as “high-risk” under Annex III and Recital 61 of Regulation (EU) 2024/1689[10]. This literature emphasises that the Act preserves the final decision-making authority of the human judge as a non-negotiable safeguard, while imposing documentation, human-oversight, and conformity-assessment obligations on providers and deployers of qualifying systems.

2.4 Indian Scholarship and Institutional Literature

Indian academic and policy literature on judicial AI is comparatively nascent but growing quickly, catalysed by the Supreme Court’s White Paper on AI and the Judiciary and by the publication of the Draft AI Regulations, 2026. Recent legal commentary traces the institutional history of SUPACE, SUVAS, LegRAA, and TERES through the three phases of the e-Courts Mission Mode Project, and situates the 2026 Draft Regulations as India’s first binding, national attempt at judicial AI governance — distinct from the advisory posture reflected in individual High Court policies, such as the Kerala High Court’s guidelines on the Adalat AI transcription tool and the Gujarat High Court’s policy barring AI use in decision-making and judgment drafting[11].

2.5 Ethical and Jurisprudential Literature

A final strand, drawing on Hartian and Dworkinian accounts of judicial reasoning, interrogates whether adjudication is reducible to a rule-application exercise amenable to computation, or whether it necessarily involves an irreducible element of moral and contextual judgment that resists algorithmic replication[12]. This literature supplies the “assistive/determinative” distinction and the concept of “meaningful human control” that structure the analytical framework adopted in this paper.

2.6 Research Gap

The reviewed literature reveals that due-process critique, bias studies, comparative regulatory analysis, Indian institutional description, and jurisprudential theory have developed largely as parallel tracks, rarely integrated into a single, doctrinally grounded framework calibrated to Indian constitutional law. This paper addresses that gap by undertaking a systematic constitutional analysis of India’s 2026 Draft AI Regulations against the comparative benchmarks of Loomis and the EU AI Act.

RESEARCH METHODOLOGY

This paper adopts a doctrinal, analytical, and comparative research methodology. Primary constitutional text and Supreme Court jurisprudence on due process, equality, and judicial independence (Articles 14, 21, 50, 124, 141, 217, and 235) are analysed alongside leading comparative case law, principally State v. Loomis. Primary regulatory instruments — the text and recitals of the EU AI Act, 2024, and the Supreme Court of India’s Draft Regulations for the Use of Artificial Intelligence in Courts, 2026 — are examined provision-by-provision. Secondary sources, including institutional press releases, the Supreme Court’s White Paper on AI and the Judiciary, National Judicial Data Grid statistics, and peer-reviewed and legal-affairs commentary, are used to construct an evidence-based account of AI deployment in Indian courts. The paper’s analytical method is explicitly comparative: each jurisdiction surveyed is assessed against a common set of constitutional values — transparency, non-discrimination, accountability, and preservation of independent human judgment — which together form the evaluative framework applied throughout.

ARTIFICIAL INTELLIGENCE IN JUDICIAL SYSTEMS: A COMPARATIVE SURVEY

4.1 The United States: COMPAS and State v. Loomis

In February 2013, Eric Loomis was arrested in Wisconsin in connection with a drive-by shooting; he pleaded guilty to lesser charges of eluding an officer and operating a vehicle without the owner’s consent. At sentencing, the trial judge relied in part on a COMPAS risk score, developed by the private company Northpointe (now Equivant), which classified Loomis as a high risk to the community and contributed to a six-year custodial sentence[13]. Loomis challenged the reliance on COMPAS on the ground that its proprietary, trade-secret methodology denied him the ability to test the scientific validity of the score, thereby violating his right to due process and to an individualised sentence.

The Wisconsin Supreme Court upheld the use of COMPAS at sentencing, but circumscribed that use: the tool could not determine whether an offender was incarcerated or the length of a sentence, had to be accompanied by an independent judicial rationale, and any presentence report containing a COMPAS score was required to carry a five-part warning — disclosing the tool’s proprietary opacity, the absence of cross-validation for the Wisconsin population, documented concerns about racial disproportionality in outcomes, and the need for continuous re-norming[14]. The United States Supreme Court denied certiorari in 2017, leaving the constitutional question of algorithmic opacity formally unresolved at the federal level[15]. Independent investigative analysis has since suggested that COMPAS-type tools may classify certain racial groups as higher-risk at disproportionate rates relative to actual reoffending outcomes, a finding that continues to animate the wider “algorithmic bias” literature[16].

4.2 The European Union: A Risk-Based Regulatory Model

The European Union’s Artificial Intelligence Act (Regulation (EU) 2024/1689) entered into force on 1 August 2024 and establishes a four-tier, risk-based classification of AI systems: unacceptable risk (prohibited outright), high risk (subject to stringent obligations), limited risk (subject to transparency duties), and minimal risk (largely unregulated)[17]. Recital 61 and Annex III of the Regulation classify as high-risk any AI system intended to be used by, or on behalf of, a judicial authority to research and interpret facts and law, and to apply law to a concrete set of facts — as well as systems used by alternative dispute resolution bodies whose outcomes produce binding legal effects[18]. Crucially, the Regulation provides that AI tools may support judicial decision-making power but must never replace it: final decision-making authority must remain human. High-risk systems are subject to obligations of risk assessment, data governance, technical documentation, human oversight, and conformity assessment before they may be placed into judicial service; purely ancillary administrative functions, such as anonymisation of judgments or internal staff communication, fall outside the high-risk classification[19].

Figure 1: The EU AI Act’s risk-based pyramid and the position of judicial AI systems within it.

4.3 The United Kingdom: The HART Custody Tool

The Durham Constabulary’s Harm Assessment Risk Tool (HART) has been used, in a policing rather than strictly judicial capacity, to inform custody decisions by predicting the likelihood of an individual’s future offending[20]. Although HART operates at the law-enforcement rather than adjudicative stage, its deployment illustrates the same underlying due-process and transparency tensions identified in the American and Indian contexts, and it is frequently cited in comparative literature as a cautionary case study of the risks of feeding demographically skewed historical data into a predictive tool with downstream consequences for liberty.

4.4 China and Estonia: Divergent Experiments

China’s “Smart Court” and “Xiao Zhi” systems represent a more determinative model of judicial AI, assisting judges with case-law retrieval, sentencing-consistency checks, and, in certain pilot contexts, generating draft rulings for judicial confirmation[21] — a model that sits closer to the determinative end of the assistive-determinative spectrum than either the American or European approaches. Estonia, by contrast, has publicly explored a strictly bounded “robot judge” concept confined to the adjudication of small-value contractual disputes below a defined monetary threshold, with a guaranteed right of appeal to a human judge — an example of a narrowly tailored, low-stakes determinative application accompanied by a structural human-oversight safeguard.

ARTIFICIAL INTELLIGENCE IN THE INDIAN JUDICIARY: INSTITUTIONAL ARCHITECTURE

India’s judicial AI experience must be understood against a backdrop of severe and chronic docket congestion. The Supreme Court’s own pendency stood at 87,115 cases as of July 2025, while nationwide pendency across all court tiers exceeded 5.4 crore matters[22].

Figure 2: Pendency of cases across the Supreme Court, High Courts, and District/Taluka courts of India (Source: NJDG; Supreme Court of India, data as of July 2025).

5.1 The e-Courts Mission Mode Project

India’s engagement with judicial technology has proceeded through three deliberate phases. Phase I (2007–2015) built foundational digital infrastructure. Phase II (2015–2023) delivered the Case Information System 3.0 and the National Judicial Data Grid (NJDG), a real-time, nationwide case-tracking database. Phase III (2023–2027), backed by an outlay of approximately ₹7,210 crore — of which ₹53.57 crore is earmarked specifically for AI and blockchain integration — has introduced the current generation of AI tools[23].

Figure 3: Timeline of AI integration in the Indian judiciary, from the e-Courts Mission Mode Project (2007) to the Draft AI Regulations (2026).

5.2 Current AI Tools

• SUVAS (Supreme Court Vidhik Anuvaad Software): a domain-specific neural machine translation system launched on Constitution Day, 26 November 2019, supporting bi-directional translation between English and nineteen Indian languages and having translated over 36,000 Supreme Court judgments as of 2026.
• SUPACE (Supreme Court Portal for Assistance in Court Efficiency): assists judges by organising the factual matrix of a case and identifying relevant precedents, without generating recommendations; as of January 2026 its use was confined to select criminal matters before the Delhi and Bombay High Courts, with the Supreme Court itself identifying hardware constraints (high-grade GPU/TPU dependency) — rather than algorithmic limitation — as the principal barrier to scaled deployment.
• LegRAA (Legal Research Analysis Assistant): a generative-AI-based research tool that analyses documents and extracts relevant legal references, operating strictly as a research aid with no role in recommending outcomes or drafting judgments autonomously.
• TERES and ASR-SHRUTI/PANINI: AI-enabled real-time transcription and dictation tools, initially deployed for Constitution Bench proceedings and progressively extended to regular hearings.

The foregoing account of India’s AI tool ecosystem is drawn from institutional press releases and legal-affairs reporting[24].

5.3 The Draft Regulations for Use of Artificial Intelligence in Courts, 2026

Published on 3 June 2026, India’s Draft AI Regulations represent the country’s first comprehensive, binding, national framework for judicial AI governance, applicable across the Supreme Court, all High Courts, and all tribunals and statutory commissions performing adjudicatory functions[25]. Its core governance principles provide that AI must remain strictly assistive and subservient to judicial authority; that final authority on law, fact, and justice vests exclusively in judicial officers; that AI must actively avoid discrimination; that opaque and unexplainable systems face heightened scrutiny; and that accountability for AI use rests personally on the officer deploying the tool. Permissible uses — case management, transcription, translation, legal research and summarisation, and administrative analytics — require prior written approval from nominated officers. Absolutely prohibited uses include algorithmic adjudication without mandatory human review, risk-scoring for bail or recidivism, outcome prediction, AI-enabled surveillance of court users, and any use compromising the confidentiality of judicial deliberations. The Draft Regulations establish an apex governance body at the Supreme Court, standing committees, and court-level AI Committees with dedicated AI Secretariats at every tier of the judicial hierarchy.

This institutional design reflects a conscious and structurally significant choice: unlike the American experience, where COMPAS-type tools entered sentencing through executive/correctional channels with limited ex-ante judicial-branch regulatory control, and unlike the more determinative Chinese model, India’s approach compartmentalises AI functions and forecloses, at the regulatory level, precisely the category of use (recidivism and outcome risk-scoring) that generated the constitutional controversy in Loomis.

CONSTITUTIONAL AND ETHICAL ANALYSIS

6.1 The Opacity (“Black Box”) Problem and Article 21

Many AI systems, particularly those built on proprietary machine-learning models, do not disclose the internal logic by which an output is generated. In Loomis, the defendant was denied access to the weighting methodology of the COMPAS algorithm on the ground of trade-secret protection[26]. Transposed to the Indian constitutional context, this raises the question whether an accused’s right to be informed of, and to meaningfully contest, the basis of a judicial or quasi-judicial determination can survive algorithmic opacity, given the Supreme Court’s consistent insistence — traceable to the expansive post-Maneka Gandhi reading of Article 21 — that a decision affecting life or liberty must be accompanied by reasons capable of judicial review[27]. A wholly opaque risk score, even if used only to “assist” a judge, risks converting a formally reasoned judicial order into a substantively unreasoned one if the human decision-maker cannot articulate why the AI output was accepted, discounted, or overridden.

6.2 The Bias and Discrimination Problem and Article 14

Empirical audits of deployed risk-assessment tools have suggested that algorithmic outputs may replicate or amplify historical patterns of discrimination embedded in training data[28]. In the Indian context, where caste-, religion-, and community-based disparities in criminal justice outcomes are independently documented, the prospect of AI systems trained on historically skewed conviction, bail, or sentencing data raises the spectre of a constitutionally impermissible violation of Article 14’s equality guarantee, dressed in the neutral vocabulary of statistical prediction. The Draft AI Regulations’ express prohibition on recidivism and bail risk-scoring can be read as a pre-emptive constitutional safeguard against precisely this risk, foreclosing the Loomis-type controversy before it can arise in India[29].

6.3 The Accountability and Separation-of-Powers Problem

Article 50 of the Constitution enjoins separation of the judiciary from the executive, and Articles 124, 217, and 235 vest the power of adjudication and superintendence exclusively in constitutionally appointed judges[30]. Where an AI system materially shapes — even without formally issuing — a judicial outcome, questions arise as to who bears responsibility for an erroneous or unjust result: the presiding judge, the system’s developer, the procuring judicial administration, or no one at all. The doctrine of judicial immunity, historically premised on the accountability of an identifiable human decision-maker exercising independent judgment, sits uneasily with a decisional process substantially informed by non-accountable software. The Draft AI Regulations’ provision that accountability for AI use rests personally on the deploying officer is a doctrinally significant response to this problem, though its practical efficacy will depend on whether judicial officers possess the technical literacy to meaningfully exercise — and be held to account for — that oversight function.

6.4 The Legitimacy, Efficiency, and Access-to-Justice Problem

India’s judiciary faces genuine and pressing docket congestion, and AI-assisted efficiency gains carry real potential to expand access to justice for litigants who currently wait years for adjudication. Institutional reporting has associated the introduction of technology-enabled case management with measurable improvements in disposal rates in at least some High Courts[31]. Yet the pursuit of efficiency cannot be permitted to erode the deliberative, empathetic, and context-sensitive qualities that are said to distinguish judicial reasoning from administrative processing. This tension — between the constitutionally sound imperative of timely justice under Article 21 and the equally constitutional imperative of individualised, reasoned adjudication — forms the ethical crux of this paper’s inquiry.

6.5 Comparative Positioning

Positioning the jurisdictions surveyed in Part 4 along a single assistive-to-determinative spectrum illustrates the range of regulatory choices available and clarifies where India’s current and proposed architecture sits relative to its peers.

Figure 4: Comparative positioning of select jurisdictions on the AI judicial-assistance spectrum.

India’s tools — SUPACE, SUVAS, and LegRAA — cluster firmly toward the assistive end of the spectrum, a positioning reinforced, and now formally entrenched, by the Draft AI Regulations’ express prohibition on determinative uses such as recidivism scoring. This is a materially more conservative regulatory posture than the American COMPAS experience and arguably more protective of due-process values than the current Chinese model, though it remains to be seen whether this conservatism will be sustained as case-management pressures intensify and technical capability expands.

FINDINGS AND A PROPOSED REGULATORY FRAMEWORK

7.1 Summary of Findings

1. Opacity is the single most constitutionally significant risk factor identified across jurisdictions: wherever an AI system’s reasoning cannot be disclosed and tested, the due-process guarantee under Article 21 is placed under strain, regardless of whether the system is formally labelled “assistive.”
2. The assistive/determinative distinction, while conceptually useful, is not self-executing in practice: a nominally assistive tool (such as a precedent-retrieval system) can become functionally determinative if judicial officers, under docket pressure, defer to its outputs without independent verification — a risk the literature terms “automation bias.”
3. India’s regulatory posture, as reflected in the Draft AI Regulations, 2026, is comparatively conservative and pre-emptively forecloses the specific category of use (recidivism/bail risk-scoring) that produced the Loomis controversy in the United States, representing a doctrinally sound response informed by comparative learning.
4. Nonetheless, the Draft Regulations leave significant implementation questions unresolved — particularly regarding the technical AI-literacy of judicial officers tasked with personal accountability for AI use, the mechanics of algorithmic audit for bias, and the enforceability of disclosure obligations vis-à-vis litigants.
5. The European Union’s risk-based classification model offers a transferable regulatory design principle — calibrating obligations to the severity of potential rights impact — that India’s binary permitted/prohibited structure could usefully absorb in more granular form.

7.2 A Proposed Graduated Framework for India

Building on the comparative and doctrinal analysis above, this paper proposes a four-tier graduated regulatory framework for AI in the Indian judiciary, structured around escalating obligations of transparency and human oversight calibrated to the potential rights impact of each category of use.

Figure 5: Proposed graduated regulatory framework for AI in the Indian judiciary, synthesising the EU AI Act’s risk-based model with India’s Draft AI Regulations, 2026 and constitutional due-process standards.

Tier I (Assistive) captures translation, transcription, and administrative analytics functions such as SUVAS and TERES, requiring minimal additional oversight beyond standard IT-security and accuracy audits. Tier II (Research Support) captures precedent- and factual-matrix-retrieval tools such as SUPACE and LegRAA, and proposes a mandatory disclosure obligation — informing parties that AI-assisted research tools were used in preparing the case file, without requiring disclosure of the underlying algorithmic architecture. Tier III (Decision-Support) would capture any future tool materially influencing an outcome-relevant determination such as bail or sentencing, and mandates a reasoned, documented human override mechanism and a full audit trail, mirroring both the EU AI Act’s high-risk obligations[32] and the personal-accountability principle already embedded in India’s Draft Regulations. Tier IV (Prohibited) restates and reinforces the Draft Regulations’ existing prohibition on autonomous adjudication, recidivism and risk-scoring, unreviewed outcome prediction, and AI-enabled surveillance of court users.

7.3 Discussion

The proposed framework is deliberately more granular than the Draft Regulations’ current permitted/prohibited binary, in order to give judicial administrators clearer, tier-specific compliance guidance and to make explicit the escalating disclosure obligations that due process arguably already implies under Article 21, even where not yet expressly codified. The framework does not purport to resolve every implementation difficulty identified in Part 7.1 — in particular, the question of how algorithmic bias audits are to be technically conducted for closed-source or vendor-proprietary tools such as those potentially procured for SUPACE’s future scaled deployment remains a matter for further empirical and technical research, appropriately falling outside the scope of a primarily doctrinal paper such as this one.

CONCLUSION AND RECOMMENDATIONS

This paper has examined the legal, ethical, and constitutional implications of Artificial Intelligence’s growing role in judicial decision-making, situating India’s institutional experience — SUPACE, SUVAS, LegRAA, TERES, and the Draft Regulations for the Use of Artificial Intelligence in Courts, 2026 — within a comparative framework anchored by the United States’ due-process jurisprudence in State v. Loomis and the European Union’s risk-based classification of judicial AI under the AI Act. The central finding is that India’s current regulatory trajectory is comparatively conservative and constitutionally well-directed, having pre-emptively prohibited the specific category of algorithmic use (recidivism and bail risk-scoring) that generated sustained due-process controversy in the American context. However, the analysis also reveals that important implementation questions — judicial AI-literacy, technical audit mechanisms for bias, and the precise contours of disclosure obligations to litigants — remain unresolved.

On the basis of this analysis, the paper offers the following recommendations:

1. The finalisation of India’s Draft AI Regulations should incorporate a graduated, tier-based disclosure and oversight obligation, of the kind proposed in Part 7.2, in place of — or as an elaboration upon — the current binary permitted/prohibited structure.
2. Mandatory AI-literacy training should be instituted for judicial officers at all tiers, as a precondition to the exercise of the “personal accountability” obligation the Draft Regulations already impose.
3. An independent, technically resourced audit mechanism — potentially housed within the proposed Apex Body and Court-level AI Committees — should be empowered to conduct periodic bias and accuracy audits of deployed tools, with findings made public in aggregate form consistent with judicial confidentiality requirements.
4. Litigants should be entitled, as a matter of codified right rather than administrative discretion, to be informed when an AI research or case-management tool has been used in the preparation of their matter.
5. Future empirical research should track disposal-rate and error-rate outcomes across AI-assisted and non-AI-assisted benches, once SUPACE’s deployment expands beyond its current pilot scope, to test — rather than assume — the efficiency case for further AI integration.

Ultimately, this paper argues that AI’s legitimate constitutional role in Indian adjudication lies in relieving the administrative and research burdens that presently obstruct timely justice — not in displacing the irreducibly human exercise of judicial reasoning that Articles 14, 21, and 235 of the Constitution presuppose. The measure of success for India’s judicial AI project will not be the sophistication of its algorithms, but its fidelity to the constitutional value it was built to serve: a fair hearing, reasoned and given by an accountable human judge.

REFERENCES

1. The Constitution of India, 1950, arts. 14, 21, 50, 124, 141, 217, 235.
2. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act), OJ L, 2024.
3. Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (published 3 June 2026).
4. Supreme Court of India, Centre for Research and Planning, White Paper on Artificial Intelligence and the Judiciary.
5. High Court of Kerala, Official Memorandum on Usage of Adalat AI.
6. High Court of Gujarat, Policy Barring AI Use in Judicial Decision-Making and Judgment Drafting.

7. State v. Loomis, 881 N.W.2d 749 (Wis. 2016), cert. denied, 137 S. Ct. 2290 (2017).
8. Maneka Gandhi v. Union of India, (1978) 1 SCC 248.

9. H.L.A. Hart, The Concept of Law (Oxford University Press, 3rd edn.).
10. Ronald Dworkin, Law’s Empire (Harvard University Press).
11. Richard Susskind, Online Courts and the Future of Justice (Oxford University Press).

12. Algorithmic Due Process: Mistaken Accountability and Attribution in State v. Loomis, Harvard Journal of Law & Technology Digest (2017).
13. State v. Loomis, Recent Case Comment, 130 Harvard Law Review (2017).
14. How the Wisconsin Supreme Court Failed to Protect Due Process Rights in State v. Loomis, North Carolina Journal of Law & Technology.
15. Judicial AI and the EU AI Act, in The Cambridge Handbook of AI and Technologies in Courts (Cambridge University Press, 2026).
16. AI and the Justice System: Approach from the EU AI Act, Asian Journal of Social Science Studies, vol. 10, no. 4 (2025).
17. Regulating Artificial Intelligence in Indian Judiciary: From Institutional Experimentation to a National Framework, Live Law Articles (2026).
18. Artificial Intelligence in the Indian Judiciary: SUPACE, SUVAS, and the Limits of Assistive Automation, Dr. Syama Prasad Mookerjee Research Foundation (2026).

19. Press Information Bureau, Government of India, Use of AI in Supreme Court Case Management (Ministry of Law and Justice, Rajya Sabha reply).
20. Press Information Bureau, Government of India, From Digitisation to Intelligence: How AI is Enhancing Access to Justice in India (2026).
21. National Judicial Data Grid (NJDG), Case Pendency Statistics, Department of Justice, Government of India.
22. European Commission, Directorate-General for Communications Networks, Content and Technology, High-Level Summary of the AI Act.
23. ProPublica, Machine Bias: There’s Software Used across the Country to Predict Future Criminals, and It’s Biased against Blacks (investigative report on COMPAS).
24. How 5 Judicial AI Solutions Are Proving Skeptics Wrong in India, ICTworks (2026).
25. Can Artificial Intelligence (AI) Revolutionize India’s Judiciary System?, IndiaAI.

26. Loomis v. Wisconsin, Wikipedia, summary of certiorari proceedings before the United States Supreme Court.
27. Wisconsin Court: Warning Labels Are Needed for Scores Rating Defendants’ Risk of Future Crime, ProPublica News Report (2020).
28. State v. Loomis: Can an Algorithm Decide a Prison Sentence?, LegalClarity (2025).
29. Integrating AI in India’s Judiciary and Law Enforcement, NextIAS Current Affairs (2025).
30. Artificial Intelligence in Judiciary, Drishti IAS Daily News Analysis.

[1]The Constitution of India, 1950, arts. 14, 21; see also Maneka Gandhi v. Union of India, (1978) 1 SCC 248 (expansive reading of Article 21 to require fair, just and reasonable procedure).

[2]State v. Loomis, 881 N.W.2d 749 (Wis. 2016), cert. denied, 137 S. Ct. 2290 (2017).

[3]Regulation (EU) 2024/1689, Recital 61 and Annex III, para. 8; European Commission, High-Level Summary of the AI Act.

[4]How 5 Judicial AI Solutions Are Proving Skeptics Wrong in India, ICTworks (2026).

[5]Press Information Bureau, Government of India, Use of AI in Supreme Court Case Management.

[6]Press Information Bureau, Government of India, From Digitisation to Intelligence: How AI is Enhancing Access to Justice in India (2026).

[7]Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026; discussed in Regulating Artificial Intelligence in Indian Judiciary: From Institutional Experimentation to a National Framework, Live Law (2026).

[8]Algorithmic Due Process: Mistaken Accountability and Attribution in State v. Loomis, Harvard Journal of Law & Technology Digest (2017); State v. Loomis, Recent Case Comment, 130 Harv. L. Rev. (2017).

[9]ProPublica, Machine Bias: There’s Software Used across the Country to Predict Future Criminals, and It’s Biased against Blacks; Wisconsin Court: Warning Labels Are Needed for Scores Rating Defendants’ Risk of Future Crime, ProPublica (2020).

[10]Judicial AI and the EU AI Act, in The Cambridge Handbook of AI and Technologies in Courts (Cambridge University Press, 2026); AI and the Justice System: Approach from the EU AI Act, Asian J. Soc. Sci. Studies, vol. 10, no. 4 (2025).

[11]Artificial Intelligence in the Indian Judiciary: SUPACE, SUVAS, and the Limits of Assistive Automation, Dr. Syama Prasad Mookerjee Research Foundation (2026); High Court of Kerala, Official Memorandum on Usage of Adalat AI; High Court of Gujarat, Policy Barring AI Use in Judicial Decision-Making and Judgment Drafting.

[12]H.L.A. Hart, The Concept of Law (Oxford Univ. Press, 3rd edn.); Ronald Dworkin, Law’s Empire (Harvard Univ. Press); Richard Susskind, Online Courts and the Future of Justice (Oxford Univ. Press).

[13]State v. Loomis: Can an Algorithm Decide a Prison Sentence?, LegalClarity (2025); State v. Loomis, 881 N.W.2d 749, 753–754 (Wis. 2016).

[14]Wisconsin Court: Warning Labels Are Needed for Scores Rating Defendants’ Risk of Future Crime, ProPublica (2020).

[15]Loomis v. Wisconsin, cert. denied, 137 S. Ct. 2290 (2017).

[16]ProPublica, Machine Bias: There’s Software Used across the Country to Predict Future Criminals, and It’s Biased against Blacks.

[17]European Commission, High-Level Summary of the AI Act.

[18]Regulation (EU) 2024/1689, Recital 61.

[19]Regulation (EU) 2024/1689, Recital 61, para. 5; Judicial AI and the EU AI Act, in The Cambridge Handbook of AI and Technologies in Courts (2026).

[20]Artificial Intelligence in Judiciary, Drishti IAS Daily News Analysis (comparative note on UK’s HART).

[21]Artificial Intelligence in Judiciary, Drishti IAS Daily News Analysis.

[22]Artificial Intelligence in the Indian Judiciary: SUPACE, SUVAS, and the Limits of Assistive Automation, Dr. Syama Prasad Mookerjee Research Foundation (2026), citing National Judicial Data Grid (NJDG) data.

[23]Integrating AI in India’s Judiciary and Law Enforcement, NextIAS Current Affairs (2025); Regulating Artificial Intelligence in Indian Judiciary, Live Law (2026).

[24]Press Information Bureau, From Digitisation to Intelligence: How AI is Enhancing Access to Justice in India (2026); How 5 Judicial AI Solutions Are Proving Skeptics Wrong in India, ICTworks (2026); Artificial Intelligence in the Indian Judiciary, Dr. Syama Prasad Mookerjee Research Foundation (2026).

[25]Regulating Artificial Intelligence in Indian Judiciary: From Institutional Experimentation to a National Framework, Live Law (2026), discussing Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026.

[26]State v. Loomis, 881 N.W.2d at 761–763.

[27]Maneka Gandhi v. Union of India, (1978) 1 SCC 248.

[28]ProPublica, Machine Bias: There’s Software Used across the Country to Predict Future Criminals, and It’s Biased against Blacks.

[29]Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026, discussed in Regulating Artificial Intelligence in Indian Judiciary, Live Law (2026).

[30]The Constitution of India, 1950, arts. 50, 124, 217, 235.

[31]Can Artificial Intelligence (AI) Revolutionize India’s Judiciary System?, IndiaAI (documenting the Madras High Court’s 114% case clearance rate in 2022).

[32]Regulation (EU) 2024/1689, Recital 61.