Published On: October 14, 2026
Authored By: Simran Kumari
Asian Law College
Abstract
Artificial intelligence is invading the realms of legal research, writing, and administrative work, but its use in adjudicating matters is proving to be a double-edged sword. Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd. (2026), which deals with the issue of AI hallucinations by authorities, is a prime example. This paper discusses the judgment against the backdrop of the Supreme Court of India’s 2025 White Paper and the Draft AI Regulations, 2026, along with the guidance provided by UNESCO, CEPEJ, and the United Kingdom, to formulate policies on judicial authenticity, accountability, and transparency, and proposes human oversight mechanisms and operational safeguards.[1]
I. Introduction
Generative artificial intelligence has advanced from assisting legal research to arguing, referencing authorities, and creating reasoning. However, its use raises the question of what effect it has on the judicial process in general.[2]
The Supreme Court of India, in Pooja Ramesh Singh v. J&K Bank Ltd., delivered on 2 July 2026, provided an answer to the question of the use of generative AI in adjudication. By annulling the decisions of the National Company Law Tribunal and the National Company Law Appellate Tribunal, which rested on non-existent, AI-generated data, the Supreme Court left room for the use of artificial intelligence as a reasoning tool but imposed a requirement of human supervision.[3]
Written at the birth of Indian law on AI, as reflected in the Supreme Court’s White Paper on Artificial Intelligence and Judiciary, 2025, the Draft AI Regulations, 2026, and the guidelines developed by organisations such as UNESCO and CEPEJ, this paper analyses the developments surrounding Pooja Ramesh Singh v. J&K Bank Ltd. In the author’s opinion, the decision of the Supreme Court goes beyond hallucination to the very heart of the judicial process itself.[4][5][6]
II. From Judicial Assistance to Judicial Dependence
Technology has been a significant ally to courts for years through e-filing, digitisation, and translation; artificial intelligence (AI) differs in that it involves intellectual functions. The Supreme Court White Paper correctly observes that contemporary AI can “understand” language, recognise patterns, draw inferences, and generate analyses.[7]
This raises the issue of a spectrum, with administrative tools such as transcription at the lower end and adjudicative functions at the higher end. AI that engages in activities such as providing reasoning or options for court decisions differs little from adjudication.
The 2026 Draft Regulations correctly prohibit adjudication by AI, but less attention has been given to the extent to which AI can assist a human decision-maker. The issue is one of proximity to adjudication: if an AI system helps select statutes, case law, and arguments, it will shape the outcome to a significant extent.[8]
III. Pooja Ramesh Singh: Hallucination and the Verification Problem
The immediate concern in Pooja Ramesh Singh is hallucination by AI: the production of fake information that masquerades as real.[9]
The Court found several citations that appear to be judicial orders but have no basis in reality. The concern is compounded because precedent derives from authenticity and authority. An assertion that masquerades as a judicial order does not acquire the force of law merely by virtue of its format. Verification is imperative, but verification is always post hoc. The damage, if any, has already been done.[10]
The SC White Paper 2025 and the UK’s 2025 Judicial AI Guidance correctly identify hallucination as a risk and emphasise the need for accuracy as well as the role of judges. These twin facts suggest that India is not in an altogether different conceptual space from the rest of the world. The reliance on generative AI carries risks that are universal, even if the contexts may differ.[11]
Accordingly, AI should always play the role of an enabler of discovery rather than a purveyor of authenticity. A citation needs to be located in a reliable source; its very existence needs to be verified.
IV. Reasoned Judgments and Reviewability
The risk of AI-generated reasoning goes beyond erroneous citations. Judgments must be informative so that litigants understand the rationale for the decision and so that appellate courts can review it.
If the reasoning process is materially informed by an AI, how can a reviewing court evaluate the reliability or provenance of that information?
While the 2026 Draft Regulations stress the importance of human oversight, auditing, and monitoring, “human involvement” can be minimal. If a judge relies on the output of an AI without evaluating or verifying it, the judge is “in the loop” but may have relinquished responsibility for reasoning and decision-making.[12]
India’s framework should require “meaningful” scrutiny: questioning, verifying, contextualising, and potentially rejecting the information provided by an AI, and providing reasons for doing so. Human oversight must be substantive, not merely illustrative, lest it become a recipe for obfuscation and organisational capture.
V. Bias, Equality and Judicial Independence
Hallucination is about false information, while bias arises from AI’s capacity to reproduce or enhance the patterns in the data used for its training or design.[13]
This raises issues of constitutional law, as Article 14’s insistence on equality and non-arbitrariness in State action holds true even when technology is involved. A “neutral” system can produce unequal results if the data and design are not neutral.[14]
The Supreme Court White Paper points out that bias is one of the most serious risks of AI in the field of law. The 2026 Draft Regulations have introduced safeguards with regard to data quality, data variety, and discriminatory bias. The UNDP’s 2026 India study also identified shortcomings in addressing fairness, equality, transparency, and judicial integrity, and recommends rights-based risk analysis and monitoring.[15][16]
Judicial independence is a further constitutional concern, as AI systems may create informational environments that make particular authorities or arguments more accessible and thus indirectly coercive.
As per the CEPEJ framework, judicial independence should be interpreted as encompassing not just the separation of the judge from external factors but also the autonomy of the decision-making process itself.[17]
VI. Open Justice and Transparency
AI can aid open justice through its capacity to translate and to facilitate access to information, but it can also lead to unjust outcomes if its role in a process is not disclosed.
The 2026 Draft Regulations and the SC White Paper rightly propose mechanisms for governance, audits, and safeguards. As UNESCO and UK guidance also recommend, India too should mandate transparency and human oversight in AI use.[18]
The rule should distinguish between uses of AI that are incidental to a process and those that substantially contribute to an outcome, with more disclosure required in the latter cases.
VII. Human-in-the-Loop: Presence Is Not Control
“Human-in-the-loop” is the preferred safeguard against automated adjudication, but its legal value is determined by the scope of human agency.[19]
The UK guidance does not prohibit the use of AI but requires that judicial office-holders take responsibility for its use, and it warns that AI may involve risks of hallucination, bias, and confidentiality breaches. Similarly, UNESCO emphasises the need for substantive, rather than token, human oversight.[20]
The Draft Regulations similarly adopt the principle of ultimate human responsibility for the acts of AI. The critical question in this context is whether the onus is on the judge to ensure that there is no error in the AI’s recommendations, or whether responsibility is allocated only when an error has occurred.[21]
The allocation of responsibility without a concomitant duty to verify would create accountability without substance. Governance norms should therefore impose a higher standard of scrutiny when the final decision-making is automated.
VIII. Authenticity: A New Dimension of Judicial Integrity
What makes a judgment genuinely judicial when its production has been assisted by artificial intelligence?
This problem is not just about accuracy. A judgment may be factually and legally correct but not contain the intellectual input of the judge in its creation. Authorship is formally stated, but substantive authorship is more difficult to establish if large parts of the judgment have been generated by AI and accepted without question.
I suggest that judicial authenticity is a component of judicial integrity: that is, the intellectual input of the decision-maker should be evident.
This does not mean that AI-assisted judgments are prohibited. Many courts may wish to use AI for legal research, document organisation and management, language translation, and administrative tasks. The issue is not one of human-written versus machine-written judgments, but one of intellectual responsibility. The relevant international organisations, including UNESCO, CEPEJ, and the UK Judicial Office, all require that AI be used as an assistant, with appropriate levels of human supervision and responsibility.
The proposed Judicial Authenticity Principle is therefore: where AI is used in the adjudicative process, the final formulations in the judgment must be capable of being substantively attributable to the human decision-maker, with due regard for the level of assistance provided by AI.
IX. Is India’s 2026 Framework Sufficient?
India’s regulatory practice is noteworthy for its rapid development: while the Supreme Court White Paper provides the conceptual and comparative basis, the 2026 Draft Regulations elaborate on the institutional governance of AI.[22]
The framework mostly aligns with global trends. Indeed, UNESCO’s emphasis on human rights, transparency, and accountability, CEPEJ’s focus on fundamental rights and a responsible judiciary, the UK’s practical guidance on the judiciary’s role, and the UNDP’s India-specific analysis of rights impact on governance design all contribute to a comprehensive view of AI governance.[23]
However, the convergence of these trends does not automatically ensure adequacy.
Three questions arise at this juncture. First, does “human responsibility” mean a responsibility of judges and courts to independently verify every single AI-assisted decision, or only a duty to exercise caution when an AI tool is used? Second, does transparency imply only the documentation of the use of AI, or the possibility of meaningful audit, thus requiring specific disclosure when its use is detected? Third, are audits not a formality if judicial organisations lack the technical expertise to assess these systems?
Thus, while the Draft Regulations should be viewed as a positive step towards responsible AI, their transformation into a detailed and accessible set of requirements is crucial to their impact.[24]
X. Conclusion: From Hallucination to Accountability
Pooja Ramesh Singh underlined the risk that AI errors create for judicial integrity. International guidelines and India’s Draft Regulations are both right to focus on the importance of human oversight, rights, transparency, and judicial independence.[25]
However, it is not enough simply to state principles. To truly ensure rights, there must be a commitment to independent auditing, provenance of AI assistance, and judicial control over decision-making.
Most fundamentally, there needs to be a presumption of judicial authenticity. In other words, the reasoning in a particular judgment needs to be genuinely the product of the judge who is authorised to deliver it. While a judge may rely on the support and research provided by various forms of artificial intelligence, the judge must not become dependent on them to the point where his or her own reasoning capacity is bypassed or overridden.
References
[1] Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., 2026 INSC 668, Civil Appeal No. 11950 of 2025, ¶ 1 (India July 2, 2026).
[2] Supreme Court of India, Centre for Research & Planning, White Paper on Artificial Intelligence and Judiciary (Nov. 2025), https://www.sci.gov.in/reports/ (last visited Aug. 13, 2026).
[3] Pooja Ramesh Singh, 2026 INSC 668, ¶¶ 1, 7, 18–21.
[4] Supreme Court of India, Centre for Research & Planning, White Paper on Artificial Intelligence and Judiciary (Nov. 2025), https://www.sci.gov.in/reports/ (last visited Aug. 13, 2026).
[5] Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (draft 2026).
[6] UNESCO, Guidelines for the Use of AI Systems in Courts and Tribunals (2025), https://www.unesco.org/en/artificial-intelligence/rule-law (last visited Aug. 13, 2026).
[7] Supreme Court of India, Centre for Research & Planning, White Paper on Artificial Intelligence and Judiciary (Nov. 2025), https://www.sci.gov.in/reports/ (last visited Aug. 13, 2026).
[8] Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (draft 2026), regs. 3(a), 19–20.
[9] Pooja Ramesh Singh, 2026 INSC 668, ¶¶ 1, 7.
[10] Id. ¶¶ 1, 7–8.
[11] Courts & Tribunals Judiciary, Artificial Intelligence (AI)—Judicial Guidance (Oct. 31, 2025), https://www.judiciary.uk/guidance-and-resources/artificial-intelligence-ai-judicial-guidance-october-2025/ (last visited Aug. 13, 2026).
[12] Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (draft 2026), regs. 4, 8–9.
[13] Supreme Court of India, Centre for Research & Planning, White Paper on Artificial Intelligence and Judiciary (Nov. 2025), https://www.sci.gov.in/reports/ (last visited Aug. 13, 2026).
[14] INDIA CONST. art. 14.
[15] Supreme Court of India, Centre for Research & Planning, White Paper on Artificial Intelligence and Judiciary (Nov. 2025), https://www.sci.gov.in/reports/ (last visited Aug. 13, 2026).
[16] United Nations Development Programme India, DAKSH & Digital Futures Lab, AI for Justice: Ethical, Fair and Robust Adoption in India’s Courts (2026), https://www.undp.org/india/ai-justice-ethical-fair-and-robust-adoption-indias-courts (last visited Aug. 13, 2026).
[17] European Commission for the Efficiency of Justice, European Ethical Charter on the Use of Artificial Intelligence in Judicial Systems and Their Environment (2018), https://www.coe.int/en/web/cepej/cepej-european-ethical-charter-on-the-use-of-artificial-intelligence-ai-in-judicial-systems-and-their-environment (last visited Aug. 13, 2026).
[18] Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (draft 2026), regs. 7–9, 37–39, 43–45.
[19] Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (draft 2026), reg. 3(zb).
[20] Courts & Tribunals Judiciary, Artificial Intelligence (AI)—Judicial Guidance (Oct. 31, 2025), https://www.judiciary.uk/guidance-and-resources/artificial-intelligence-ai-judicial-guidance-october-2025/ (last visited Aug. 13, 2026).
[21] Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (draft 2026), regs. 4, 8.
[22] Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (draft 2026), regs. 7–9, 37–39, 43–45.
[23] UNESCO, Guidelines for the Use of AI Systems in Courts and Tribunals (2025), https://www.unesco.org/en/artificial-intelligence/rule-law (last visited Aug. 13, 2026).
[24] Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (draft 2026).
[25] Pooja Ramesh Singh, 2026 INSC 668, ¶¶ 1, 7.
