Should AI be allowed in judicial decision-making? Examine the ethical, constitutional, and practical dimensions of regulating artificial intelligence in Indian courts.

The Supreme Court's draft Regulations for Use of AI in Courts, 2026 answer this by drawing a firm line: AI may assist, never adjudicate, and must remain "strictly subservient to human judgment and judicial authority" [1]. This calibration — assistive yes, decisional no — is the defensible position.

The regulatory line drawn

• Permitted: legal research, citation verification, transcription, translation, cause lists, scheduling and case management [1].
• Prohibited: judgments, sentencing, bail eligibility, assessing witness credibility, risk scoring, profiling of parties, and surveillance of judicial officers [1].

Ethical dimension

• Algorithmic bias: models trained on historical records can entrench caste, class and gender patterns; the bar on profiling shields vulnerable witnesses from statistical stereotyping.
• Automated bail risk-scoring would harden existing disadvantage in a system where undertrials form roughly three-fourths of the prison population [3].
• Judicial accountability is non-delegable — the CJI has assured that AI will never overpower judicial decision-making [4].

Constitutional dimension

• Article 21 guarantees a fair trial and personal liberty; Article 14 bars arbitrariness — both are compromised if liberty turns on an unauditable score.
• The right to a reasoned order under natural justice explains the ban on "opaque or unexplainable" systems, effectively mandating explainable AI [1].
• Mandatory disclosure of AI-assisted filings extends existing candour obligations of counsel to algorithmic tools [1].

Practical dimension

• Hallucination risk: a Bench censured a trial court for citing non-existent AI-generated judgments, terming it misconduct rather than mere error [1].
• Institutionally, a full-time apex body at the Supreme Court is proposed to approve and standardise court AI [1], while the reconstituted SC AI Committee steers adoption across the higher and subordinate judiciary [2].
• Uneven digital infrastructure across High Courts and district courts remains the real compliance challenge.

Thus AI should be admitted as an efficiency tool and denied any adjudicatory role. A human-in-the-loop design, explainability audits and phased capacity-building can convert AI into a genuine instrument of speedy justice under Article 21, reducing pendency without diluting the judge's constitutional conscience.