Author
Sneha Awasthi is a 4th-year student pursuing the 5-year LL.B. programme at Chhatrapati Shahu Ji Maharaj University, Kanpur.
The Draft AI Regulations for Courts 2026, issued by the Supreme Court’s AI Committee, seek to rethink judicial independence in India.
Introduction
At a time when the Indian judiciary is already badly strained by institutional pressures, AI is making its mark in the courts. However, judicial processes are not the same as usual administrative procedures. A court is more than a processor of information. It evaluates facts, listens to different points of view, and provides reasons that impact liberty, property, and legal rights. Whether AI can speed up courts without compromising judicial independence remains the question.
The Supreme Court’s Draft Regulations for Use of Artificial Intelligence in Courts, 2026, seek to answer this question. The Draft, issued by the Supreme Court’s AI Committee on 3 June 2026, aims to govern the use of AI in the Supreme Court, High Courts and other adjudicatory bodies. It professes to uphold human primacy, transparency, accountability, data protection and judicial independence. That makes the governance of judicial AI a constitutional issue.[^1]
The Draft offers a solid foundation for responsible judicial AI, especially through its absolute prohibitions on algorithmic decision-making. But formal control does not lead to meaningful human judgment. The more challenging question is whether judges and court personnel, already under intense institutional constraints, can become dependent on AI-generated summaries, research and analysis over time.
The Transition from Digital Courts to Intelligent Courts
The adoption of AI in the legal field in India has been a gradual process. The approved outlay of ₹7,210 crore for Phase III takes the project towards cloud infrastructure, open APIs, artificial intelligence and machine learning.[^2]
SPACE helps in legal research, extraction and retrieval of relevant precedents. SUVAS uses machine translation to make Supreme Court judgments available in Indian languages. ASR-SHRUTI is developed to provide voice-to-text transcription, while PANINI is developed to provide linguistic processing of Indian legal language. There is an emerging emphasis on the use of generative tools for controlled judicial analysis in systems like LegRAA.[^3]
The Draft Regulations: Human Primacy as the Starting Point
According to Regulation 4, the use of AI shall remain subordinate to human judgment and judicial authority, while Regulation 5 subjects its use to the Constitution, existing laws, natural justice and judicial conduct principles.[^4]
AI cannot make a judgment, order, finding of fact or law. It cannot sentence or adjudicate without human involvement. Also, it is not allowed to be used for risk scoring, such as recidivism prediction, bail eligibility or party and witness credibility. The prohibition extends to the use of undisclosed or unexplainable AI in situations involving rights and personal liberty. AI-generated content cannot be presented as standalone evidence without disclosing its true nature.[^5]
In Vijay Ghanshyam Gadiya v. Union of India, the Supreme Court found the underlying order to be based on non-existent authorities, fake citations and propositions that were seemingly AI-hallucinated, and set aside a customs penalty. The Court reiterated that AI can be used to aid the adjudication process, but cannot take its place.[^6]
The Draft thus establishes an important institutional boundary. But the question that has yet to be settled is not what is written in the prohibition, but whether human verification will remain truly independent when AI becomes of daily judicial practice. AI might affect the reasoning process by determining what a judge reads first, which authorities seem relevant, or how a large record is condensed, without officially exercising judicial authority. The nature of human involvement with the result is therefore the real measure of human primacy, rather than just the presence of a human decision-maker at the end of the chain.
The Draft is not, however, exclusively based on prohibitions alone. It also establishes an institutional framework to regulate the use of AI. It sets up an Apex Body at the Supreme Court level, Standing AI Committees and AI Secretariats at the High Courts, requires prior approval for the use of AI systems, mandates technical and ethical impact assessments, allows for testing in controlled environments, and institutes annual audits, an AI Register and an AI Incident Database. These mechanisms are key because human primacy cannot rely solely on the personal caution of a judge. It also needs institutions that are able to detect an unreliable system and prevent its use from continuing.[^7]
The challenge is how these protections will be implemented. Approval without a proper understanding of the system being approved is only meaningful if the approving authority has sufficient technical knowledge. Likewise, an annual audit might uncover a problem after it has affected many cases. Incidents should be reported and followed up through the Incident Database if it is to be useful in identifying recurring failures. The Draft sets up the framework of accountability, but its effectiveness depends on the independence, expertise and resources of the institutions to which it is to be applied, whether national or regional.
Opportunities don’t wait. Neither should you.
Join 1 Lakh+ law students connected with Lexibal and stay updated with internships, opportunities, competitions and important updates.
Join WhatsApp ChannelConstitutional Concerns: When Human Oversight Becomes Formal
Articles 14 and 21 demand a fair, non-arbitrary and meaningful judicial process. A reasoned opinion should be based on the judge’s evaluation of the evidence presented to the court. The problem is that AI might seep into the judge’s thinking process, and that is dangerous.
This is an example of what is called automation bias. The worry isn’t that a judge will blindly accept an AI-generated output. The fact that an output is delivered in an organised and authoritative manner can lead it to be given more authority than it has earned, especially if the judge has limited time and a heavy workload. So human oversight must involve the ability and willingness to challenge, rather than merely the authority to disapprove of, the output.
When assessing flight risk, recidivism or credibility, historical patterns can translate into seemingly neutral evaluations of individuals. An algorithm can potentially amplify social and institutional bias if these biases exist in the underlying data. Although Article 14 does not say that the use of technology gives arbitrariness its legitimacy, Article 21 requires more attention when liberty is involved.
According to Annex III, AI systems designed to support judicial authorities in research, interpretation or application of the law are considered high-risk systems. The Act does not simply leave oversight to the final decision-maker. Article 14 requires human oversight within the operation of the system, including monitoring and the ability to intervene or override outputs.[^8]
When a judge accepts an AI-generated bench note and signs an order, the judge has formally exercised judicial authority. However, if the judge does not have enough time or information to check the machine’s analysis, the safeguard is more of a formality.
Data Protection and Institutional Inequality
Personal information, including material and documents of a confidential or proprietary nature connected with current matters, are part of court records. Regulation 47 requires compliance with the Digital Personal Data Protection Act, 2023 and the Information Technology Act, 2000.[^9]
While compliance on paper is essential, it isn’t enough to resolve all questions raised by judicial AI. Personal information, confidential communications and commercially sensitive information may be found in court records. If personal information is fed into an AI system, there are concerns about where the data is housed, who has access to it, whether it is used to improve an external model, and how long it is stored. These concerns are exacerbated if the system is run by third-party technology providers.
While the infrastructure and technical staffing may be better in the Supreme Court and High Courts, many other courts remain constrained by basic technology. A regulatory model that mandates advanced AI assessments, which are absent in many institutions, could lead to disparate protection across courts.
Towards Meaningful Human Primacy
Courts should maintain a clear record of the use of AI in judicial practice. Such a record should include the kind of system used, the purpose of its use and the stage of the proceeding at which it was used. If the use of AI directly assists in the preparation of a document or in analysing the record, it should be possible to trace this. This would make it easier to review an error relating to AI after a dispute occurs.
Second, approved judicial AI systems should undergo independent technical and legal audits routinely. Assessment should include accuracy, bias, data sources, and the occurrence of hallucinations across different types of cases. The consequences of such incidents should also be recorded in an incident database. If material errors are repeated, the system must be reviewed, have its approval withdrawn, or be suspended.
Third, judicial officers and registry personnel need the ability to learn about, practise and work with AI. They should be apprised of how such errors are created in generative systems, how summaries can leave out key information, and how citations that may appear authoritative must be verified independently. Training must focus on developing judicial judgment, not technological dependence.[^10]
Conclusion
The Draft Regulations for 2026 include an important principle: Artificial intelligence can support the judiciary, but it cannot replace human judgment. Its absolute prohibitions are a strong feature of the Draft. The real challenge is to keep human primacy alive in the midst of ever more ubiquitous use of AI in research, in case management and preparation.
India must therefore not only restrict the decisions AI cannot make, but also regulate how humans interact with AI before making decisions. That requires meaningful oversight and independent auditing, reliable data governance, institutional capacity and appropriate training. The true test of judicial AI is whether the technology enhances the justice system while maintaining the principle that justice is administered by persons. A machine can be used to organise information in a court. The duty to consider that information, reach a decision and provide reasons for it must remain human.
Footnotes
[^1]: Supreme Court of India, Draft Regulations for Use of Artificial Intelligence in Courts, 2026 (June 3, 2026), regs. 4–5, available in the draft text published by the Supreme Court AI Committee.
[^2]: Press Information Bureau, Government of India, Cabinet Approves eCourts Phase III for 4 Years (Sept. 13, 2023), noting the approved financial outlay of ₹7,210 crore and the planned use of AI, machine learning, OCR and NLP in the e-Courts ecosystem.
[^3]: Ministry of Law & Justice, Government of India, Use of Artificial Intelligence Tools in Judicial System, Lok Sabha Starred Question No. 147 (Dec. 16, 2022) (discussing SPACE and SUVAS and their intended use in the judicial system).
[^4]: Supreme Court of India, supra note 1, regs. 4–5.
[^5]: Id., Reg. 20. The Draft Regulations describe the prohibitions in Regulation 20 as “absolute and non-derogable” and prohibit, inter alia, judicial outcomes based solely on AI-generated information or analysis, AI-based risk scoring and certain opaque or undisclosed uses of AI.
[^6]: Vijay Ghanshyam Gadiya v. Union of India, 2026 INSC 947 (Sept. 2, 2026). The Supreme Court found that authorities relied upon in the impugned customs order were either non-existent, supported by fake citations, or did not support the propositions attributed to them.
[^7]: Supreme Court of India, supra note 1, regs. 19, 21–42, 49–50. The Draft provides for institutional mechanisms including approval requirements, impact assessments, controlled testing, audits and incident-related mechanisms.
[^8]: 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), arts. 6(2), 14, Annex III, 2024 O.J. (L 168) 1. Annex III specifically includes AI systems intended to assist judicial authorities in researching and interpreting facts and law and applying law to concrete facts among high-risk systems.
[^9]: Supreme Court of India, supra note 1, reg. 47. Regulation 47 addresses the application of relevant data-protection and information-technology laws to judicial AI.
[^10]: Supreme Court of India, supra note 1, regs. 49–50. The Draft specifically provides for capacity building, training and a repository of best practices relating to AI incidents.
