Reducing Pendency, Preventing Misuse: AI in India’s Judicial System

INTRODUCTION. India’s courts face persistent pendency and misuse of procedural levers. I present a governance-first, judgment-neutral registry-intake assistant that operates on filing metadata only, without predicting case outcomes or recommending sanctions. METHODS. Intake is framed as administrative triage on structured signals (e.g., filing subtype, party/advocate counts, completeness flags, duplicate-hint hashes, and time-to-first-action). A lightweight attention mechanism (logistic regression) surfaces at-risk filings for clerk review; no end-to-end adjudication modeling is used. Evaluation uses a synthetic, CPU-only 12-week workload with weekly clustering and 95% confidence intervals (CIs). RESULTS. The assistant improved front-end handling rate and reduced processing time at intake; early-adjournment risk was flagged sooner for clerk attention. Evidence accuracy (flags-only) reached 0.89 (95% CI: 0.88–0.91). Ablations showed incremental gains from (i) registry-rules intake logic, (ii) administrative signals only, and (iii) learned attention; the learned piece remained bounded to metadata. DISCUSSION. The design aligns with EU AI Act operator duties (risk management, data governance, logging, human oversight) and India’s e-Courts Phase III rails, with Digital Personal Data Protection Act, 2023 (DPDP) compliance (data minimization, role-based access, retention). Post-market monitoring and incident reporting are built into the workflow. CONCLUSION. A judgment-neutral, metadata-only intake assistant can improve administrative efficiency while avoiding adjudication predictions and automated sanctions, offering a practical approach to responsible AI in justice.