AI Use by the Judiciary: SC’s Draft AI Regulations, 2026

Why in the News?

The Supreme Court released the Draft Regulations for Use of Artificial Intelligence in Courts, 2026 last month, inviting public comments till July 15. The draft permits AI for administrative and research functions in courts but places an absolute, non-derogable bar on any AI role in decisions affecting bail, recidivism (a critical metric used to measure the effectiveness of the justice and rehabilitation systems.), witness credibility, or personal liberty.

What does the Draft Regulations permit AI to do in courts?

1. Administrative and assistive functions: AI use is permitted for case management, transcription, translation, legal research, document summarisation, accessibility, and court administration.
2. Approval requirement: Every permitted use requires prior written approval from the Apex Body for the Supreme Court, or the AI Committee of the concerned High Court or tribunal.
3. Human supervision: Officers nominated by the court must supervise and verify AI-assisted outputs before use.
4. Scope boundary: Permission covers efficiency-enhancing functions only. It does not extend to any function that produces or contributes to a judicial outcome.

Why has the SC opted for a staggered, court-wise implementation instead of a uniform rollout?

1. SC-specific notification: Provisions apply to the Supreme Court only from a date notified by the Chief Justice of India.
2. High Court autonomy: Provisions for High Courts and the courts and tribunals under their jurisdiction come into force separately, on dates notified by the respective High Court Chief Justice.
3. Provision-wise phasing: Different provisions can be brought into force on different dates within the same court.
4. Rationale: Phasing allows each court to adopt AI at a pace suited to its own infrastructure, caseload, and readiness.

Why is human judicial authority made non-negotiable in adjudicative outcomes?

1. Categorical bar on algorithmic outcomes: No judicial outcome can be reached through algorithmic decision-making alone, or solely on the basis of AI-generated information.
2. Determinative human authority: Human judicial authority is determinative in all adjudicative decisions, regardless of AI input.
3. Advisory-only role: Where AI is used anywhere in a decision-making process, its role is only advisory.
4. Independent evaluation mandate: Any AI-assisted input is subject to independent human judicial evaluation before use.

What functions has the SC placed beyond regulatory reach altogether, and why?

1. Risk scoring barred: AI cannot be used for ‘risk scoring’ to assess flight risk.
2. Recidivism prediction barred: AI cannot be used to predict recidivism.
3. Bail eligibility barred: AI cannot be used to evaluate bail eligibility.
4. Witness credibility barred: AI cannot be used to determine the credibility of witnesses.
5. Profiling barred: AI cannot be used to predict, profile, or infer the future conduct or behaviour of parties, accused persons, witnesses, or legal representatives.
6. Undisclosed AI evidence barred: AI-generated output cannot be submitted as independent evidence without full disclosure of its AI-generated character.
7. Blackbox AI barred in liberty matters: Unexplainable AI systems cannot be used in matters affecting personal liberty.
8. Non-derogable status: These prohibitions are absolute. No authority can permit them later under the Regulations.

Does the disclosure mechanism for litigants adequately safeguard their right to know?

1. Material assistance trigger: Litigants must be informed only when an AI tool “materially assists” case management, document analysis, or judicial administration.
2. Timely and accessible disclosure: Disclosure to litigants and their counsel must be made in a timely and accessible manner.
3. Threshold-based, not blanket disclosure: Litigants are not informed of every instance of AI use in their case, only instances that meet the material assistance standard.
4. Undefined threshold: The Regulations do not define what constitutes “material assistance,” leaving the disclosure trigger to case-by-case determination by courts.

What institutional architecture will govern AI use in courts?

1. Apex Body: An Apex Body at the Supreme Court will set minimum mandatory standards for AI systems and issue implementation guidelines.
2. Composition: The Apex Body comprises sitting Supreme Court and High Court judges, an official of the Ministry of Electronics and Information Technology, and experts in finance and cybersecurity.
3. Specialised committees: The Apex Body will function through five specialised committees.
4. Court-level AI Committees: The Supreme Court and each High Court will constitute their own AI Committees, backed by an AI Secretariat.
5. Dedicated research body: The Centre of Research and Excellence on Artificial Intelligence (CoRE-AI) will evaluate AI tools and track technological developments to support the Apex Body.

How are private AI vendors regulated to prevent capture of judicial data and infrastructure?

1. Prior written approval: Private companies can supply AI tools only with written approval from the relevant court authority.
2. Mandatory contract terms: Vendor agreements must include a mandatory list of contract terms set out by the Regulations.
3. Data ownership and access: Contracts must specify ownership of, and access rights to, court data and AI outputs.
4. Bar on sensitive data use: Vendors are barred from using sensitive judicial data.
5. No unauthorised model training: Vendors cannot retain or fine-tune models using court data without the AI Committee’s written approval.
6. IP restriction: Vendors cannot claim exclusive intellectual property rights over tools built substantially using public resources.

Conclusion

The Draft Regulations construct a two-tier framework for judicial AI: broad permission for administrative efficiency, and an absolute prohibition on AI’s role in outcome-determinative and liberty-affecting functions. This boundary, not the list of permitted uses, is the framework’s operative safeguard against algorithmic opacity compromising due process. The undefined “material assistance” threshold for litigant disclosure remains its weakest link, leaving courts significant discretion over what litigants get to know. Effective implementation will depend on how the Apex Body and CoRE-AI operationalise this boundary as AI adoption scales across courts.

PYQ Relevance

[UPSC 2024] Explain the reasons for the growth of public interest litigation in India. As a result of it, has the Indian Supreme Court emerged as the world’s most powerful judiciary?”

Linkage: The PYQ discusses expansion of judicial power through institutional self-assertion. The Draft AI Regulations are another instance of the SC using its institutional authority to self-regulate its own processes.