By Bhavin Patel and Khushi Singh
Welcome to The Bridge! India's regulators affect markets, govern disputes, and test the rule of law through their orders, regulations, and policy implementation actions. Yet the quality of this work rarely gets the scrutiny it deserves. The Bridge is TrustBridge's window into India's regulatory state: how it decides, how it regulates, and how it could do both better.
Introduction
In August 2024, a bench of the National Company Law Tribunal (“NCLT”) passed a judgment in which it relied on AI-generated precedent to support its findings. Interestingly, the AI-generated precedent slipped under the radar and found mention even in the judgment of the appellate tribunal, the National Company Law Appellate Tribunal (“NCLAT”). Following an appeal against this judgment, the Supreme Court in Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd1 (“Pooja Ramesh Singh”) expressed the need to control when and where to apply AI in the adjudicatory process. For our analysis of this AI-generated error and proposed solutions, see our previous post here.
In its decision, the Supreme Court acknowledged that regulations for this purpose were already under deliberation, referring to the Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (the “Draft Regulations”), issued on 3rd June 2026 and open for comments until 15th July 2026. It is unclear what law authorises the Supreme Court to issue these ‘regulations’. We address this issue, as well as other concerns on the Supreme Court’s jurisdiction in this matter in a later section of this post.
The Draft Regulations have been prepared by the Supreme Court’s ‘Artificial Intelligence Committee’. Reportedly, the Committee comprised a Supreme Court judge and four judges from various High Courts.2 However, we have not noticed any record of formal consultation with lower courts or tribunals before the release of the Draft Regulations.
While we submitted formal comments (available here) during this consultation window, our analysis here goes beyond those comments.We suggest that while the Draft Regulations are well-intentioned, they extend into domains where the Supreme Court’s jurisdiction is questionable. Even in establishing rules for the use of AI in administrative processes within their own court, they remain lacking in both clarity and precision on a few aspects. In this post, we analyse the specific areas where the Draft Regulations may stumble in their aspirations.
What do the Draft Regulations cover?
The overall approach of the Draft Regulations towards the use of AI in the judiciary is covered in certain principles provided under Chapter II. These include human primacy, fairness and non-discrimination, transparency and explainability, final accountability resting on judicial officers as the user, auditability and oversight, proportionality and purpose limitation, and inclusivity and accessibility.
The Draft Regulations do three things substantially:
(i) Lay down permissible and prohibited uses of AI: Permissible uses include automated transcription, judgment translation, legal research retrieval, docket scheduling and case filing assistance. Prohibited uses include independent adjudication without human oversight, future behaviour profiling, and predictive risk-scoring;
(ii) Create several policy-making and administrative authorities such as the “Apex Body” and “AI Committee”, and outline their functions; and
(iii) Provide rules for the judiciary’s engagement with private entities for AI tool procurement.
Apart from this, the Draft Regulations specify requirements for matters such as audits, reviews and incident reporting.
Our assessment of the Draft Regulations
Potential overreach: Trying to do too much with little jurisdiction
The Draft Regulations apply to “Supreme Court of India, High Courts, and all Courts including the “Tribunals and statutory Commissions” performing adjudicatory functions, within the territory of India”.3 The AI committees of the High Courts can “issue supplementary directions tailored to local needs” but only as long as they are consistent with the minimum mandatory standards issued by the Supreme Court’s Apex Body.4
The Draft Regulations leave the terms “Tribunals” and “statutory Commissions” undefined.5 In general usage, the Draft Regulations imply an intent to cover statutory commissions such as the National Human Rights Commission, the Securities and Exchange Board of India, or the Competition Commission of India. Similarly, “Tribunals” likely encompasses entities such as the NCLT or the Securities Appellate Tribunal.
Because these entities are statutorily designed to remain independent and oversee their own administrative frameworks, it remains unclear whether the regulations intend to exclude them by targeting a specific subset of adjudicatory bodies. If the regulations are meant to exclude these autonomous bodies, the terms “Tribunals and statutory Commissions” must be explicitly defined. Otherwise, if they are included, the Supreme Court’s stipulation over how the use of AI should be governed by these independent bodies sitting outside the judiciary becomes problematic. What is clear is that the regulations purport to bind all other courts. But the authority to do even this is questionable.
The Supreme Court does not have power to lay down administrative rules for any other court or tribunal in India. Dave 2026 argues that Articles 141 and 145 cannot be stretched to interpret that the Supreme Court has the authority to dictate prohibitions on the other courts.6 Article 141 binds lower courts on legal precedent made by the Supreme Court, not administrative matters, and Article 145 only allows the Supreme Court to make house-keeping rules for itself. Conversely, supervisory authority over subordinate courts and tribunals remains with the High Courts.7
The Draft Regulations lay down enforceable rules that even go beyond administrative matters. For instance, the regulations create a substantive rule that no private entity can claim exclusive IPR over tools developed primarily using “judicial data or public resources”.8 Regardless of one’s perspective on whether judges create law, it is difficult to validate the Supreme Court expanding its role from interpreting legal frameworks to enacting entirely new substantive positive legislation.
All of this is not to say that the Supreme Court should do nothing. In other jurisdictions as well, judicial bodies are leading efforts to control the use of AI within the adjudicatory realm. A prominent example is the Courts and Tribunals Judiciary of England and Wales, which published its AI Guidance for Judicial Office Holders.9 Unlike the Draft Regulations in India, these frameworks are typically framed as “guidance” or “guides”. While these are not binding, they are directed at all judicial officers. This may be less contentious in the context of England and Wales, since they are issued in the name of the Chief Justice and Senior President of Tribunals who, between them, head the judiciary and tribunals respectively.
Similarly, the judiciary in Singapore introduced the Guide on the Use of Generative Artificial Intelligence Tools by Court Users10 (“Singapore’s Guide”), which is applicable to all courts. Rather than just the supreme court acting alone, every court in Singapore adopted an identical set of guidelines. This uniformity appears to arise from mutual consensus rather than an effort to compel compliance from lower courts.
While the Supreme Court may arguably direct other courts and tribunals to frame guidelines for the use of AI for their respective proceedings within a specified timeframe, the ultimate responsibility of drafting and implementing these rules vests with these entities. Alternatively, if the Supreme Court is considered the proper authority to control use of AI across all judicial bodies, then its jurisdiction must be expanded by the Parliament in accordance with Article 138 of the Constitution.11
Missing middle: “High-Risk AI Tools”
The Draft Regulations only clearly lay down permissible uses and prohibited uses.12 There is also, however, a middle category of “High-Risk AI Tools” that the regulations make scattered references to.13 The language of the regulations suggest that “applications involving higher levels of risk” are allowed, “subject to correspondingly heightened safeguards…”14 A risk-based, graded approach to AI is appropriate, as it aligns with the principle of “innovation over restraint” established in the regulations. However, the regulations neither define the criteria for “High-Risk AI Tools” nor provide what heightened safeguards will apply.
This guidance can come from the Supreme Court’s “Apex Body” which is responsible for “standard-setting and policy development”.15 Ordinarily, handing off such broad authority is a case of excessive delegation. But here, the Court lacks a legislative mandate to begin with, and so it does not have anything to hand off to a delegatee. Standard-setting and policy development are functions typically reserved for Parliament. The court creating a body to exercise legislative powers raises separation of powers concerns.
Consequences for non-compliant judicial officers?
The Draft Regulations clarify that accountability for all decisions made with AI assistance will rest exclusively upon the officer using the AI.16 They also establish that responsibility for submitting fabricated, false, misleading, or inaccurate AI-generated content to the court rests with the individual who presents it. Nonetheless, the regulations fail to outline the specific consequences that court officers will face in the event of a breach.17 They merely state that if an AI system is employed for prohibited purposes, the AI Committee is empowered to suspend it.18 Contrast this with Singapore’s Guide which clarifies that failure to comply with the Guide can lead to imposition of costs and disciplinary action.19
Others
There are a few other aspects of the Draft Regulations which seem incomplete. For instance, while AI-generated output or information used in any Court needs to be verified, the officers using the AI tool are granted wide discretion to dispense with the mandatory verification of AI-generated output by recording their reasons in writing.20 The regulation should provide some guidelines or a certain threshold for waiver of the verification requirement. Providing these objective boundary conditions will support consistent decision-making and offer clearer guidance for processing officers. Additionally, although the Draft regulations assign responsibilities to the AI Secretariat and AI Committee that necessitate technical proficiency, they fail to require the inclusion of individuals with a technical background.21 Furthermore, the framework of a “presumption in favour of responsible AI” ought to be restricted to low-risk tasks rather than being applied universally.22
Conclusion
The impulse of the Supreme Court in Pooja Ramesh Singh was right in that it seeks to balance technological integration with cautious application within the judiciary. The Draft Regulations also declare “innovation over restraint” as one of the principles.23 Nevertheless, the Draft Regulations attempt to control too much, with questionable authority to do so.
As the court rightly points out, “[t]he real success is, however, not in the making of the Rule or Regulation, but to be found in the power of the will of the Bar as well as the Bench, to harness this science and apply it with care and caution”.24 Rather than immediately enforcing strict “regulations”, the court could instead consider initiating the process by providing “guidance”.
The burden of control of AI in adjudication does not solely rest on the judiciary. While courts should establish administrative guidelines detailing the permitted and prohibited uses of AI, the task of driving behavioural change, training, and providing guidance to advocates is better suited for Bar Councils. This perspective aligns with the Supreme Court’s direction in Pooja Ramesh Singh, instructing the Bar Council of India to constitute a committee and deliberate on the issue of advocates submitting AI-generated material before the Court.25 England & Wales Bar Council’s note, for instance, helps lawyers in understanding the technological basis and inherent risks in the use of generative LLM systems.26
Ultimately, the judiciary’s role is to act as a steward of justice, not a legislator of technology.
- The authors are researchers at TrustBridge, and would like to thank Pratik Datta for his feedback.
CITATION
Bhavin Patel and Khushi Singh, 2026. “The Supreme Court's Draft AI Regulations: On what authority?”, The Bridge, TrustBridge Rule of Law Foundation
Thanks for reading The Bridge! Subscribe for free to receive new posts and support our work.
References
Constitution of India, Arts 138, 141, 145, 227
Courts and Tribunals Judiciary (England and Wales), ‘AI Guidance for Judicial Office Holders’ (31 October 2025)
‘CJI Surya Kant Reconstitutes Supreme Court’s Artificial Intelligence Committee; Justice PS Narasimha To Lead’ (LiveLaw, 10 December 2025) <https://www.livelaw.in/top-stories/cji-surya-kant-reconstitutes-supreme-courts-artificial-intelligence-committee-justice-ps-narasimha-to-lead-312920>
Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (Supreme Court of India, issued 3 June 2026)
England & Wales Bar Council, ‘Considerations when using ChatGPT and Generative Artificial Intelligence Software Based on Large Language Models’ (last reviewed 25 November 2025)
General Clauses Act 1897
Nipun Dave, ‘Is The Supreme Court Overstepping On Its Draft AI Regulations, 2026?’ (LiveLaw, 14 June 2026) <https://www.livelaw.in/articles/draft-ai-rules-apply-high-courts-537713>
Pooja Ramesh Singh v Jammu & Kashmir Bank Ltd, 2026 INSC 668 (Supreme Court of India)
Singapore Courts, ‘Guide on the Use of Generative Artificial Intelligence Tools by Court Users’ (effective 1 October 2024)
2026 INSC 668, para 5.
“CJI Surya Kant Reconstitutes Supreme Court's Artificial Intelligence Committee; Justice PS Narasimha To Lead” (LiveLaw.in, 10 December 2025) <https://www.livelaw.in/top-stories/cji-surya-kant-reconstitutes-supreme-courts-artificial-intelligence-committee-justice-ps-narasimha-to-lead-312920>
Draft Regulations for Use of Artificial Intelligence in Courts, 2026, Regulation 2.
Ibid, Regulation 33(3)(c).
These terms are also not defined elsewhere, such as the General Clauses Act, 1897.
Nipun Dave, “Is The Supreme Court Overstepping On Its Draft AI Regulations, 2026?” (14 June 2026) <https://www.livelaw.in/articles/draft-ai-rules-apply-high-courts-537713>
Constitution of India, Art. 227. Under Regulation 1(2)(b), each High Court can decide when these Draft Regulations take effect for itself and all courts, Tribunals, and Statutory Commissions within its jurisdiction. High Courts can also set different start dates for different parts of the regulations. However, this effectively puts the High Courts in a difficult position as not notifying the Draft Regulations for their courts may draw the ire from the Supreme Court.
Ibid, Regulation 46(9).
Courts and Tribunals Judiciary, “AI Guidance for Judicial Office Holders” (31 October 2025).
Singapore courts, “Guide on the Use of Generative Artificial Intelligence Tools by Court Users” (effective 1 October 2024).
Constitution of India, Art 138: “(1)The Supreme Court shall have such further jurisdiction and powers with respect to any of the matters in the Union List as Parliament may by law confer. (2) The Supreme Court shall have such further jurisdiction, and powers with respect to any matter as the Government of India and the Government of any State may by special agreement confer, if Parliament by law provides for the exercise of such jurisdiction and powers by the Supreme Court.”
Draft Regulations for Use of Artificial Intelligence in Courts, 2026, Regulations 19 and 20.
See ibid, Regulation 46(4)((h).
Ibid, Regulation 12(2).
Ibid, Regulation 22(1).
Ibid, Regulation 8.
Ibid, Regulation 52(1).
Ibid, Regulation 21.
Singapore courts, “Guide on the Use of Generative Artificial Intelligence Tools by Court Users” (effective 1 October 2024), clause 9.
Draft Regulations for Use of Artificial Intelligence in Courts, 2026, Regulation 8(3).
Ibid, Regulations 33(2) and 34.
Ibid, Regulations 16(1) and 17(2).
Ibid, Regulation 17.
Pooja Ramesh Singh (n 1) para 5.
Ibid, para 9.
England & Wales Bar Council, “Considerations when using ChatGPT and generative artificial intelligence software based on large language models” (last reviewed on 25 November 2025).