Madras High Court allows AI-assisted record analysis in arbitration batch, with strict guardrails
The Madras High Court has approved a controlled trial of an AI-assisted record system in arbitration matters between Gammon-OJSC Mosmetrostroy JV and Chennai Metro Rail Limited. In a common order dated January 28, 2026, Justice N Anand Venkatesh permitted the tool, "Superlaw Courts," to support document management while keeping all judgment calls with the Court.
"In essence, Superlaw Courts functions as an exceptionally organised and cautious record assistant: it prepares the papers, creates a reliable finding aid, and presents relevant excerpts on request, while remaining strictly bound to the record and leaving all legal judgment to human decision-makers."
Scope and limits set by the Court
- Record-only operation: The system works strictly on documents filed in the matter. It does not access external sources and does not invent or infer missing content. If a passage is absent, the system states that it cannot find it.
- No legal conclusions: It will not draw legal inferences, assess credibility, interpret intent, or express legal opinions. Human decision-makers remain in full control.
- Workflow assistance: It converts scans to searchable text, groups related materials, removes duplicates, and segments records into coherent excerpts rather than arbitrary page splits.
- "Sealed record room" workspace: The Court likened the digital setup to a closed record room to streamline handling of voluminous arbitral files.
- Full transparency: A separate link will log every interaction by counsel and the Court so any reader can see how and when the tool was consulted.
- Drafting boundary: AI may assist a draft order limited to facts, pleadings, evidence, and rival submissions for verification by both sides. After that stage, AI use stops; final adjudication is solely by the judge.
How the pilot will run
- The algorithm was demonstrated in open court with participation from both sides and the judge. A detailed working note was circulated and placed on record.
- Counsel for both parties agreed to a one-week trial and will report back on effectiveness.
- The matters are posted for final hearing from February 12, 2026, with a fixed schedule for submissions and replies.
- Appearances: Petitioners represented by Senior Advocate Sivanandaraj with Advocate PJ Rishikesh; Respondent represented by Advocate Harishankar.
Practical takeaways for counsel and arbitrators
- File hygiene matters: Use text-searchable PDFs, consistent exhibit labeling, and clean indexes. Good inputs improve recall and reduce errors.
- Check the logs: Your queries may be visible via the public log link. Assume your prompts and retrievals could be reviewed.
- Keep reasoning human: Use the tool to surface excerpts quickly, then build the legal analysis yourself. Don't outsource arguments.
- Verify every excerpt: Cross-check AI-surfaced passages against the original record before relying on them in court or in written submissions.
- Mind confidentiality: Treat the workspace like a sealed room. Confirm access controls and sharing protocols before uploading sensitive materials.
Why this matters
Arbitral records are often massive. A tightly bounded, record-focused assistant-used under judicial supervision-can reduce retrieval time and clerical friction without touching the merits. The Court's guardrails offer a workable template for tech adoption that respects due process and preserves the judge's role.
For official updates and orders, visit the Madras High Court.
If you're building AI literacy for litigation teams, see curated AI courses by job to standardize skills across your practice.
Your membership also unlocks:
