The Supreme Court's AI committee has released a draft of the 'Regulations for the Use of Artificial Intelligence in Courts, 2026'. Its core message is clear - AI will play only an assistant's role in the judicial process; the final authority of a decision will remain with the human (the judge). This is the first broad institutional framework of AI governance in India's judiciary.
Why in the News
The Supreme Court's AI committee released the draft regulations and sought suggestions from 3 June 2026 by notification until 20 June, then extended the date to 15 July 2026 by a notification of 16 June. The draft has come at a time when on 2 July 2026 the Supreme Court had warned about AI-generated 'misleading' (hallucinated) fake case citations - incidents surfaced of lawyers presenting fake judgment references made by AI (the Supreme Court's judgment, 2 July 2026). In July 2026 this draft was at the public-suggestions stage. In a limited check on 17 October 2026, no evidence of the final regulations being notified was found; legal analysis of September 2026 (SCC Times, 10 September) also treats them as a 'draft'. Treat the Supreme Court's notification alone as proof of the final position.
Key Facts
What the draft is: a draft of the 'Regulations for Use of AI in Courts, 2026' prepared by the Supreme Court's AI committee; comments were sought by the notification of 3 June 2026 (first until 20 June), then extended to 15 July 2026 by the notification of 16 June.
Permitted uses (11 categories): uses such as translation, legal research, precedent search, citation verification, document management, and automatic preparation of administrative formats/summonses - but for each, written approval from the apex body or the AI committee of the concerned high court/tribunal and the supervision of a named officer are mandatory. For a use outside the list, prior written permission with reasons is necessary.
Complete prohibitions (Regulation 20): the prohibitions are 'complete and non-derogable': AI training from personal data without approval; no judicial outcome from a purely algorithmic decision; no decision/sentence without human control; 'risk scoring' of bail eligibility, recidivism prediction or witness credibility is prohibited; non-transparent AI that affects freedom/rights is prohibited; predicting the future behaviour of parties is prohibited; surveillance in court premises is prohibited; AI content without disclosure is not evidence.
The accountability rule: responsibility for any content prepared with AI's help will rest entirely on the concerned officer; a technical fault, the model's limit or 'hallucination' (AI's fabricated output) cannot become a basis for escaping responsibility. AI output is advisory only; verification is mandatory.
The institutional structure: in the Supreme Court a permanent full-time apex body - two SC judges nominated by the CJI (one as chairperson) and two high court chief justices; with five specialised committees and an AI secretariat; every high court has its own AI committee. Violations of prohibitions are to be reported to the AI secretariat; the AI committee will investigate and remedy (including suspending a system).
Technical-ethical assessment: for every approved AI system a Technical and Ethical Impact Assessment is mandatory - an assessment of the architecture, training data, bias, hallucination risk and cyber security.
Why This Draft Matters
The nature of a judicial decision: justice is not only information processing, but the application of discretion, sensitivity and constitutional values. By giving 'human-in-the-loop' an institutional form, the draft protects this character of justice from technical substitution.
The credibility crisis: AI-generated fake precedents (hallucinated citations) have become a global problem; in India too they have reached the courts. Mandatory citation verification and the disclosure obligation are a direct answer to it.
The stop on risk scoring: the complete ban on tools like recidivism prediction is important - international experience (such as the COMPAS controversy) shows that such tools perpetuate historical biases and turn procedural justice into algorithmic prediction.
The ethical model of governance: this is a mixed model of 'prohibition + approval + accountability' - neither a complete ban (which would stop innovation) nor uncontrolled adoption (which raises risk). It signals the direction of India's broader AI governance framework.
Analysis and Way Forward
Three challenges lie ahead. First, implementation capacity - setting up an AI committee, secretariat and verification system in every high court is resource-intensive; the lower courts, where the backlog is the highest, have the least technical capacity. Second, the line between 'assistant' and 'decisive' can blur in practice - when a judge relies on an AI recommendation, the formal human intervention may remain only nominal; so a reasoned written verification is necessary. Third, the draft is limited to the court sphere - the use of AI in police, prosecution and jail administration is still without a comparable framework. The final regulations will come after the suggestions process is complete; at the draft stage this is a 'soft law' direction, not binding law.
Distinctions Within the Concept
Regulations versus an Act: this draft is an internal regulatory framework proposed for the judicial process by the Supreme Court's AI committee - not an Act passed by Parliament. It limits the use of AI in courts; it is different from society-wide AI regulation (such as the EU AI Act). It is also different from the 'e-Courts project' (digitisation, e-filing, virtual courts) - that is the digital foundation, this is an AI-specific ethical-procedural framework.
Mains Perspective
GS-II: the judiciary's structure, organisation and functioning; technology in justice; ethics and accountability in governance. GS-III: science and technology, governance of emerging technologies like AI. Essay/ethics (GS-IV): technical decision versus human discretion, the ethical limits of algorithmic justice.
Answer Outline: Introduction: the draft context (June-July 2026, the suggestions stage). Main Body: the 5 principles; the 11 permitted categories + prior approval; the complete prohibitions (risk scoring, non-transparent AI, surveillance); accountability on hallucination; the apex body/committees/TEIA. Balance assessment: positive - innovation allowed under supervision, judicial finality with humans; risks - implementation capacity, nominal human intervention, limited sphere. Conclusion: from soft law towards institutional maturity.
Sources
- Supreme Court AI Committee: the draft Regulations 2026 and notification (PDF)
- Supreme Court: Pooja Ramesh Singh v. J&K Bank - the judgment on AI-generated fake precedents, 2 July 2026 (PDF)
Verification note: verified by reading the primary text of the draft regulations (definitions, permitted/prohibited uses, the apex body's structure, Regulation 20). The draft was at the suggestions stage in July 2026; a limited check on 17 October 2026 found no evidence of a final notification - see the Supreme Court's notification for the final position.
Remember: the formula of the draft Regulations 2026: AI = assistant, judge = decision-maker; no use without approval, and uses like 'risk scoring' never.
