Introduction
The Indian judiciary is at a crossroads. More than 54 million cases are currently pending across courts at every level,1 and that backlog alone has made the case for technological intervention almost impossible to resist. Artificial intelligence has emerged as an obvious candidate: it can sift through enormous volumes of legal records, render judicial documents across 19 languages, and forecast how a case might unfold, all at a speed no human registry could match. But folding AI into the act of adjudication raises questions that go straight to the foundations of constitutional governance. Can an algorithm genuinely discharge a judge’s duty to provide reasons? Can machine learning ever reproduce the constitutional conscience that shaped judgments such as Maneka Gandhi v. Union of India2 or K.S. Puttaswamy v. Union of India?3 And, perhaps most importantly, what limits must be placed on AI before it stops serving the courtroom and starts running it?
These questions have gained new urgency from two institutional documents released within the last year: the Supreme Court’s November 2025 White Paper on AI and the Judiciary,4 and the Draft Regulations for Use of Artificial Intelligence in Courts, 2026, put out for public consultation on 3 June 2026.5 Taken together, these instruments mark India’s most serious attempt to establish a principled framework for judicial AI. They insist that AI must remain assistive, that final adjudicatory authority stays with judicial officers alone, and that AI systems must be built to avoid discrimination. This study puts these commitments to the test.
This paper is organised into ten sections. Section II presents the research questions. Section III reviews the existing literature on the topic. Section IV explains the methodology. Section V situates AI within the constitutional framework of India. Section VI examines the actual use of AI through SUPACE and SUVAS. Section VII offers a critical reading of the Draft AI Regulations, 2026. Section VIII examines comparative frameworks abroad. Section IX proposes a rights-based governance model with specific recommendations for its implementation. Section X concludes the study.
A. Research questions
Four questions guide this study and define its scope.
RQ1. To what extent does deploying AI in Indian judicial proceedings engage the fundamental rights guaranteed under Articles 14, 19, and 21 of the Constitution, and what are the minimum constitutional standards that should govern that deployment?
RQ2. Do the AI tools already operating in India, principally SUPACE and SUVAS, meet the constitutional requirements of transparency, non-discrimination, and procedural fairness, or do they create risks with which the Constitution should be concerned?
RQ3. Do the Draft Regulations for Use of Artificial Intelligence in Courts, 2026, supply a constitutionally adequate governance framework for judicial AI, and where do the institutional and legislative gaps remain?6
RQ4. What can India learn from comparative judicial AI frameworks, particularly those of Brazil, Singapore, and Canada, to build a rights-based model suited to its own constitutional and socio-legal conditions?7
B. Literature review
The relationship between AI and judicial governance has drawn considerable scholarly attention worldwide, although doctrinal scholarship focused specifically on India has not kept pace with the speed of institutional change. This review maps the main strands of existing work and identifies where this study fits.
The constitutional groundwork for this enquiry has already been well established. M.P. Jain’s treatise on Indian constitutional law sets out the foundational reading of Articles 14 and 21 as substantive guarantees of equality and procedural fairness, not merely formal ones.8 The landmark ruling in Maneka Gandhi v. Union of India9 and K.S. Puttaswamy v. Union of India10 carried those guarantees into the digital age, recognising that privacy and reasoned procedure are bound up with constitutional personhood itself. Upendra Baxi’s early scholarship made a related point: constitutional adjudication is irreducibly normative, which matters directly for any claim that AI can replicate judicial judgment.11
Internationally, scholarship on AI in judicial systems has passed through three phases: an early wave of enthusiasm about efficiency gains, a critical turn focused on bias and opacity, and a more recent normative-institutional phase centred on governance frameworks. The UNESCO Global Toolkit on AI and the Rule of Law for the Judiciary (2023) is the most comprehensive synthesis produced to date, treating explainability, accountability, and human oversight as non-negotiable conditions for deploying judicial AI.12 Its four-principle framework, built around transparency, interpretability, accountability, and reproducibility, supplies a technically grounded standard that this study applies directly to the Indian context.
India-specific scholarship has been growing quickly. Solanki and Pareek’s 2026 study, published in the Legal Reference Services Quarterly, offers the most recent systematic look at AI integration in the Indian judiciary and identifies hardware dependency and linguistic diversity as the structural constraints holding back SUPACE.13 The Dr. Syama Prasad Mookerjee Research Foundation’s February 2026 report on SUPACE and SUVAS provides a technically detailed account of both tools, covering their architecture, workflow limits, and equity implications.14 NITI Aayog’s National Strategy for Artificial Intelligence provides the broader policy backdrop against which any judicial AI initiative must be read.15
Three gaps stand out in the literature. First, no one has systematically mapped India’s judicial AI initiatives against the specific guarantees of Articles 14, 19, and 21. Second, given the recent promulgation of the Draft AI Regulations, 2026, their constitutional implications have not yet received sustained scholarly attention. Third, the equity question of whether AI deployment will widen the gap between metropolitan and district courts remains largely unexamined. This study considers all three aspects.
C. Methodology
This study relies on the doctrinal legal method, supplemented by comparative and normative analyses. The doctrinal approach involves a systematic reading of primary legal sources, constitutional provisions, Supreme Court and High Court judgments, legislative instruments, and subordinate regulations to identify, describe, and critically assess the rules that currently govern judicial AI in India. The primary sources examined include Articles 14, 19, and 21 of the Constitution; the line of Supreme Court judgments from Maneka Gandhi to Puttaswamy; the Draft Regulations for Use of Artificial Intelligence in Courts, 2026; the Digital Personal Data Protection Act, 2023; and the Information Technology Act, 2000.
The doctrinal core is supplemented by a comparative survey of judicial AI governance in Brazil, Singapore, and Canada. The point of the comparison is not to import foreign solutions wholesale but to identify principles that recur across jurisdictions, namely human primacy, explainability, and accountability, and then to ask how those principles translate into India’s own constitutional setting, which differs from the comparators in its entrenched fundamental rights, its basic structure doctrine, and the sheer scale of its case backlog.16
The normative element of this paper builds a proposed governance framework on the basis of the doctrinal analysis and the comparative survey. This framework is grounded in the constitutional text and existing Supreme Court jurisprudence and is designed to be implemented through ordinary legislation and judicial administration without requiring a constitutional amendment.17
This study does not rely on empirical methods such as interviews, surveys, or quantitative analysis. This choice fits the nature of the questions at stake, which are normative and constitutional rather than empirical. Future research examining how judicial officers use SUPACE in practice, and how aware litigants are that AI has touched their case, would complement the analysis offered here.18
The constitutional framework: Articles 14, 19, 21, and judicial independence
A. The right to equality and non-discrimination (Article 14)
Article 14 of the Constitution guarantees that the State shall not deny any person equality before the law or the equal protection of the laws within India’s territory.19 In State of Madras v. V.G. Row, the Supreme Court made clear that this equality is substantive, not merely formal: it requires that similarly situated persons be treated alike, and that any distinction bear a rational connection to a legitimate state purpose.20 AI systems trained on historically skewed judicial data risk encoding structural bias along the lines of caste, gender, class, and religion directly into their outputs. This risk converts existing social inequality into something that resembles algorithmic inevitability. Any AI tool whose outputs disparately affect identifiable groups, without rational justification, would sit on constitutionally shaky ground under Article 14.21
The Draft AI Regulations, 2026, try to address this by requiring AI systems to actively avoid discrimination and by subjecting opaque or unexplainable systems to heightened scrutiny.22 But a regulatory mandate on paper means little without an independent audit mechanism that can actually recall or suspend a non-compliant system. Section VI addresses this gap.
B. The right to life and personal liberty: procedural fairness under Article 21
Article 21 provides that no person shall be deprived of life or personal liberty except according to the procedure established by law.23 In Maneka Gandhi v. Union of India, the Supreme Court held that this procedure must itself be fair, just, and reasonable, and cannot be arbitrary, fanciful, or oppressive.24 This reading imports a substantive due process dimension into Article 21, and it has serious consequences for any system of AI-assisted adjudication. The duty to provide reasons is at the heart of procedural fairness. In S.N. Mukherjee v. Union of India, the Supreme Court held that the obligation to give reasons is itself a facet of the rule of law, one that keeps decision-making from sliding into arbitrariness.25 Where an AI system produces an output (a risk score, a case summary, a sentencing recommendation) without an explicable chain of reasoning, and a judicial officer adopts that output without independent scrutiny, the resulting order sits on weak constitutional footing. It fails to meet the reasoned decision-making standard embedded in Article 21.
C. Freedom of expression and access to justice (Article 19(1)(a))
Article 19(1)(a) guarantees freedom of speech and expression,26 and its relevance to judicial AI lies mainly in access to justice as part of meaningful legal participation. In Shreya Singhal v. Union of India, the Court recognised that vague or overbroad state action can chill speech.27 By analogy, AI tools that process a litigant’s case history or personal data without adequate transparency may chill participation in judicial proceedings, especially among marginalised communities already wary of algorithmic surveillance. A litigant’s right to understand the basis of a decision that affects her is an emanation of Article 19, read with Article 21.28
D. Judicial independence as a basic structure feature
Judicial independence forms part of the basic structure of the Constitution, as the Court affirmed in Minerva Mills Ltd. v. Union of India29 and reaffirmed in Supreme Court Advocates-on-Record Association v. Union of India.30 Article 50 directs the State to take steps to separate the judiciary from the executive.31 The Draft AI Regulations rightly name judicial independence as a core governance principle,32 but the risk of executive or commercial interference in how an AI system is designed, whether through biased training data or proprietary black-box algorithms supplied by private vendors, poses a structural threat that no aspirational clause can address on its own.
AI in Indian courts: SUPACE, SUVAS, and the limits of assistive automation
A. SUPACE: the case analysis tool
The Supreme Court Portal for Assistance in Courts Efficiency, or SUPACE, runs on a four-tier workflow: file indexing, a conversational chatbot interface, a logic gate that sorts extracted material into synopsis, chronology, evidence, and cited case law, and an integrated notebook where judges can gather and annotate information using voice-to-text.33 Crucially, the logic gate does not weigh merits or suggest outcomes. It simply organises the material so that a judge can reach the relevant part of the record faster. This design choice reflects a constitutionally sound instinct: the machine curates, and the judge decides.
Even so, as of early 2026, SUPACE remains experimental, deployed only in select criminal matters before judges in the Bombay and Delhi High Courts.34 The main constraint on a wider rollout is not the software itself but the hardware: deep learning models need high-grade GPUs, and procuring and maintaining them is expensive and logistically difficult across India’s scattered court infrastructure.35 That raises an equity concern of its own. The benefits of AI-assisted adjudication may end up flowing disproportionately to metropolitan courts, widening the justice gap in district and subordinate courts where most Indian litigants seek redress.
B. SUVAS: bridging the language barrier
The Supreme Court Vidhik Anuvaad Software, or SUVAS, is a domain-specific neural machine translation system trained on judicial texts to keep legal terminology accurate in context. As of 2026, it translates both ways between English and 19 Indian languages, including Hindi, Kannada, Tamil, Telugu, Punjabi, Marathi, Gujarati, Malayalam, Bengali, and Urdu.36 SUVAS works through a human-in-the-loop process: AI-generated draft translations are reviewed by trained human translators before anyone relies on them. That design is constitutionally sound in a straightforward way: it uses AI to extend human capacity rather than to replace human judgment. Of the judicial AI tools currently used in India, SUVAS is probably the most constitutionally defensible. Its domain-specific training, human oversight mechanism, and narrow, clearly bounded scope make it a template worth copying elsewhere. Section IX returns to this point when it sets out a proposed governance framework.
C. The hallucination problem and fake citations
In early 2026, the Supreme Court issued a stern warning against AI-generated fake case citations appearing in pleadings, treating their submission as potential professional misconduct.37 The phenomenon, large language models confidently inventing precedents that do not exist, exposes a real weakness in any uncritical reliance on generative AI within the legal field. A litigant’s right to be heard on the basis of accurate legal material, which Article 21 implies, is directly threatened when fabricated citations slip into the judicial record undetected. The Draft AI Regulations attempt to address this by requiring lawyers to declare their use of AI and verify its accuracy,38 but the enforcement side of that requirement remains thin.
Critical analysis of the Draft AI Regulations, 2026
A. The human primacy principle
Regulation 7 states, in unambiguous terms, that final authority over law, fact, and justice rests exclusively with judicial officers, and that AI must remain strictly assistive.39 This is constitutionally necessary. As Upendra Baxi observed, the judicial function is not merely cognitive but normative: it involves applying value-laden constitutional principles to contested facts in ways that no training dataset can fully capture.40 The human primacy principle is, in effect, how the requirement of reasoned judicial decision-making under Articles 14 and 21 gets operationalised.
B. Transparency, explainability, and the black box problem
Regulation 49 requires that opaque and unexplainable AI systems face heightened scrutiny before they are deployed in court.41 That provision confronts the deepest problem in judicial AI, the so-called black box phenomenon, where complex deep learning models produce outputs through processes that even their own designers cannot fully explain. In constitutional terms, an unexplainable AI recommendation that a court adopts without independent scrutiny undermines the reasoned decision-making requirement built into Article 21, along with the transparency dimension of Article 14.
The UNESCO Global Toolkit identifies four principles of explainable AI: transparency, interpretability, accountability, and reproducibility, treating all four as prerequisites for judicial deployment.42 The Draft AI Regulations gesture toward these principles but stop short of laying down technical standards or certification requirements for any AI system seeking judicial approval. This gap risks turning the explainability mandate into something hortatory rather than enforceable.
C. Data governance and the right to privacy
The right to privacy, recognised as fundamental under Article 21 in K.S. Puttaswamy v. Union of India,43 bears directly on judicial AI. Court proceedings generate enormous quantities of sensitive personal data: criminal histories, financial records, medical evidence, and family disputes. Regulation 38(2) of the Draft AI Regulations addresses data protection obligations,44 but it has to be read alongside the Digital Personal Data Protection Act, 2023,45 and the Information Technology Act, 2000,46 before it amounts to a workable data governance framework for judicial AI. There is still no court-specific data protection regime distinct from the general civilian regime, and this gap needs urgent legislative attention.
D. Accountability and audit
Regulations 52 and 53 establish an accountability framework that places personal responsibility for AI use on the judicial officer who employs the tool.47 This is a reasonable starting point, but it raises a practical question: can a judicial officer realistically be expected to audit the algorithmic integrity of a complex AI system while adjudicating a case? This paper argues that personal accountability needs to be backed up by institutional accountability, through an independent AI Audit Committee with the technical capacity to evaluate, certify, and, if necessary, suspend judicial AI systems.
Comparative perspectives: Brazil, Singapore, and Canada
A comparative look shows that India’s emerging framework broadly tracks global best practices while retaining certain nationally specific features. Brazil’s National Council of Justice, through Resolution 615/2025, requires human oversight for every AI-assisted judicial decision and sets up a national AI committee with real enforcement powers.48 Singapore’s Supreme Court Guide on the Use of Generative AI Tools by Court Users places ultimate responsibility for all submissions on lawyers and litigants, requires that cited cases and provisions be verified by a human, and prohibits reliance on unverified AI output. Canada’s Judicial Council, in its 2024 Guidelines for the Use of AI in Canadian Courts, insists that AI must never substitute for judicial reasoning and that transparency toward litigants is non-negotiable.49
What sets the Indian constitutional context apart is the explicit entrenchment of equality, dignity, and procedural fairness as judicially enforceable fundamental rights, alongside the basic structure doctrine’s protection of judicial independence. Those features place a more demanding constitutional standard on judicial AI governance in India than exists in jurisdictions built on parliamentary sovereignty or weaker traditions of judicial review. The Draft AI Regulations must be judged against that higher constitutional baseline, not simply against administrative best practice.
Recommendations: towards a rights-based governance framework
Building on the doctrinal analysis, literature review, and comparative survey set out above, this paper offers five recommendations for a constitutionally grounded, rights-based framework for judicial AI governance in India. They are addressed to three actors: the Parliament of India, the Supreme Court acting in its administrative capacity, and the National Judicial AI Authority proposed below.
A. Enact a dedicated judicial AI governance statute
The Draft AI Regulations, 2026, are subordinate regulatory instruments, and their enforceability and constitutional legitimacy would be considerably strengthened by a dedicated parliamentary statute, a Judicial AI Governance Act, that expressly grounds AI deployment standards in Articles 14, 19, and 21, establishes an independent National Judicial AI Authority, and creates a private right of action for litigants harmed by non-compliant AI use in adjudication.50
B. Mandatory explainability standards for all judicial AI systems
No AI system should be allowed into judicial proceedings unless it meets independently verifiable explainability standards, developed by the National Judicial AI Authority in consultation with the Bureau of Indian Standards. Deep learning systems that cannot produce an intelligible reasoning chain should be excluded outright from any decision-support role in adjudication, in keeping with Article 21’s requirement of a reasoned procedure.51
C. Establish an independent National Judicial AI Authority
An independent AI Audit Committee, comprising technical experts, judicial officers, civil society representatives, and constitutional law specialists, should have the power to certify, monitor, and suspend judicial AI systems. The Committee should publish annual transparency reports open to the public, consistent with the accountability norms already embedded in Article 14 and the right to information under Article 19(1)(a).52
D. Litigant notification, transparency, and right of challenge
Wherever AI tools have touched the processing of a litigant’s case, whether for case management, translation, or record analysis, that litigant should be told which specific tools were used, what their outputs were, and how far those outputs shaped the judicial officer’s decision. This requirement follows directly from the right to a fair hearing implicit in Articles 21 and 14, and is supported by the privacy jurisprudence of Puttaswamy.53
E. Equity in deployment
AI deployment should not deepen the existing digital divide between metropolitan and district courts. A dedicated infrastructure fund ought to ensure that AI tools, once certified, reach every level of the court system uniformly, with priority given to the district and subordinate courts where case backlogs are most severe. NITI Aayog’s National Strategy for Artificial Intelligence already provides a policy basis for this type of equity-oriented deployment.54
Conclusion
AI’s arrival in India’s courts is neither avoidable nor, by itself, something to fear. What is avoidable, and what the Constitution requires to be avoided, is the uncritical, opaque, and inequitable use of algorithmic tools that undermine the reasoned, transparent, and impartial adjudication that Articles 14, 19, and 21 demand. The Supreme Court’s White Paper and the Draft AI Regulations, 2026, are promising first steps, but they remain aspirational documents whose real constitutional value depends on whether the institutions actually follow through.
India has an opportunity here that few constitutional democracies have managed to seize: to build, from the ground up, a judicial AI governance architecture that is constitutionally grounded, technically rigorous, institutionally accountable, and equitable in its reach. The basic structure doctrine, fundamental rights jurisprudence, and a tradition of active judicial review together give India a constitutional foundation stronger than that of most comparable jurisdictions to work with on this problem. The real question is not whether AI will enter Indian courtrooms. It already has. The question is whether it enters as a servant of justice or slips its leash and becomes the master.55
*****
Footnotes
1. National Judicial Data Grid (reporting over 54 million pending cases across Indian courts as of 2026).
2. Maneka Gandhi v. Union of India, (1978) 1 SCC 248, 284 (India).
3. K.S. Puttaswamy v. Union of India, (2017) 10 SCC 1, 267 (India).
4. Supreme Court of India, White Paper on Artificial Intelligence and the Judiciary 3-5 (Nov. 2025).
5. Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts reg. 2 (2026).
6. See generally Draft Regulations, supra note 5; White Paper, supra note 4.
7. UNESCO, Global Toolkit on Artificial Intelligence and the Rule of Law for the Judiciary 56-60 (2023) (comparative survey of the Brazil, Singapore, and Canada frameworks).
8. M.P. Jain, Indian Constitutional Law 1385 (8th ed. 2018); Minerva Mills Ltd. v. Union of India, (1980) 3 SCC 625, 653 (India).
9. Maneka Gandhi, (1978) 1 SCC at 284.
10. Puttaswamy, (2017) 10 SCC at 267.
11. Upendra Baxi, The Little Done, the Vast Undone, 9 J. Indian L. Inst. 323, 330 (1967).
12. UNESCO, supra note 7, at 43.
13. Taruna Solanki & Animesh Pareek, AI and Technology in the Indian Judiciary: A Step Toward Enhancing Efficiency and Equity, 45 Legal Reference Servs. Q. 85, 91-93 (2026).
14. Dr. Syama Prasad Mookerjee Research Foundation, Artificial Intelligence in the Indian Judiciary: SUPACE, SUVAS, and the Limits of Assistive Automation 7-9 (Feb. 2026).
15. NITI Aayog, National Strategy for Artificial Intelligence 36-40 (June 2018).
16. White Paper, supra note 4, at 3-5; Draft Regulations, supra note 5, reg. 2.
17. Baxi, supra note 11, at 330; Jain, supra note 8, at 1387.
18. Maneka Gandhi, (1978) 1 SCC at 284; S.N. Mukherjee v. Union of India, (1990) 4 SCC 594, 609 (India).
19. India Const. art. 14.
20. State of Madras v. V.G. Row, (1952) SCR 597, 607 (India).
21. Indra Sawhney v. Union of India, (1992) Supp (3) SCC 217, 512 (India).
22. Draft Regulations, supra note 5, reg. 20; India Const. art. 14.
23. India Const. art. 21.
24. Maneka Gandhi, (1978) 1 SCC at 284.
25. S.N. Mukherjee v. Union of India, (1990) 4 SCC 594, 609 (India).
26. India Const. art. 19(1)(a).
27. Shreya Singhal v. Union of India, (2015) 5 SCC 1, 85 (India).
28. E. Indriasari, Y. Wahyudi, M. Taufiq & R.V. Neonbeni, Legal and Ethical Challenges in the Use of Artificial Intelligence for Judicial Decision-Making in the Technology Era, 4 Formosa J. Sci. & Tech. 1495, 1495-1506 (2025), https://doi.org/10.55927/fjst.v4i6.103.
29. Minerva Mills Ltd. v. Union of India, (1980) 3 SCC 625, 653 (India).
30. Supreme Court Advocates-on-Record Ass’n v. Union of India, (2016) 5 SCC 1, 74 (India).
31. India Const. art. 50.
32. Draft Regulations, supra note 5, reg. 7.
33. Dr. Syama Prasad Mookerjee Research Foundation, supra note 14, at 7-8.
34. White Paper, supra note 4, at 18-19.
35. J.M. Garrigus, Hardware and Software Optimizations for Deep Learning Workloads on Graphics Processing Units (2024) (Ph.D. dissertation, University of North Texas), https://doi.org/10.12794/metadc2415982.
36. White Paper, supra note 4, at 11-12.
37. Krishnadas Rajagopal, Supreme Court Warns Against AI-Generated Fake Citations, The Hindu (Mar. 15, 2026).
38. Draft Regulations, supra note 5, reg. 52.
39. Baxi, supra note 11, at 330.
40. Id..
41. Draft Regulations, supra note 5, reg. 49.
42. UNESCO, supra note 7, at 43.
43. Puttaswamy, (2017) 10 SCC at 267.
44. Draft Regulations, supra note 5, reg. 38(2).
45. The Digital Personal Data Protection Act, 2023, No. 22, Acts of Parliament, 2023, sec. 4 (India).
46. The Information Technology Act, 2000, No. 21, Acts of Parliament, 2000, secs. 43A, 72A (India).
47. Draft Regulations, supra note 5, regs. 52-53.
48. L. Liberato Junior & A.B. Nogueira, O Uso da Inteligência Artificial no Processo Judicial, 16 Lumen et Virtus e7042 (2025), https://doi.org/10.56238/levv16n51-003.
49. UNESCO, supra note 7, at 56-58.
50. Jain, supra note 8, at 1385-87.
51. S.N. Mukherjee, (1990) 4 SCC at 609; India Const. art. 21.
52. Draft Regulations, supra note 5, reg. 7; White Paper, supra note 4, at 22.
53. Puttaswamy, (2017) 10 SCC at 267.
54. NITI Aayog, National Strategy for Artificial Intelligence 36 (June 2018).
55. Law Comm’n of India, Report No. 277: Wrongful Prosecution (Miscarriage of Justice): Legal Remedies 14 (Aug. 2018).