Singapore's courts are AI-assisted. So why are lawyers still printing $35,000 worth of paper?

SINGAPORE: A recent defamation case in the Singapore courts ran up a printing bill of more than S$35,000. Not legal fees. Printing. Hard copies of hearing bundles, written submissions, affidavits and exhibits, produced and reproduced for a case where the underlying evidence spanned years of correspondence.

The case is not unusual. CNA spoke to several litigators for a recent explainer on the issue, and the consistent answer was that high printing costs are a routine feature of document-heavy litigation. "High printing costs are routine particularly in document-heavy litigation cases," said Mr Jonathan Tan, a partner in dispute resolution at Withers KhattarWong. Cases with the highest printing costs, he said, tend to involve evidence spanning a long period, emails and messages sent over several years, or financial records such as pay slips and invoices.

What makes the bill notable is not the amount alone. It is where it is being spent. Singapore's court system has run an electronic filing system since 1997, upgraded over time into the current eLitigation platform. Paper filing at a court counter is barred outright, with narrow exceptions. The Singapore Judiciary has an active memorandum of understanding with Harvey.AI. Generative AI case summarisation tools are already deployed for Small Claims Tribunal users, and AI-powered translation of court documents into Chinese, Malay and Tamil has been live since 2024. By most measures, this is one of the more digitally advanced court systems anywhere. Lawyers are still printing tens of thousands of dollars of paper per case.

01Digitisation and digital transformation are not the same thing

The gap comes down to a distinction that is simple to state and hard to close in practice. Filing a document electronically is digitisation. Making that document, and everything connected to it, genuinely searchable and instantly retrievable by everyone entitled to see it, is digital transformation. Singapore's courts have done a large amount of the first. Most of the second is still missing, and the printing bill is one visible symptom of that gap.

Electronic hearing bundles exist. Hard copies are still required for certain matters, appeals among them, according to lawyers interviewed by CNA. Evidence spanning years still has to be manually compiled, checked and physically bound before some proceedings can use it. The documents were filed digitally. The system built around them was not designed to make years of correspondence instantly queryable.

02What Trusted DataVerse would add, and how it differs from e-filing

This is the layer Trusted DataVerse is built for, and it is a different job from what eLitigation already does. e-filing gets a document into the system. DataVerse is what happens after that. It ingests documents, images and audio, corrects and validates them, and converts them into structured data that can be reasoned over rather than merely stored. Years of correspondence stop being a folder read chronologically and become a properly indexed evidentiary record.

Every text-based document processed this way under an Evidence Act certified process also carries the presumption of authenticity under Section 116A, a principle covered in an earlier TechBytes piece. Records produced through an approved process shift the burden of proof away from the party relying on them. Digitised evidence processed this way is not only more convenient. It is more defensible.

Centralised, rights-managed safekeeping is the other half of the idea. Instead of every party independently printing and transporting a hearing bundle, a case's evidentiary record could sit in one governed environment, with judiciary, counsel on both sides, and litigants each granted only the access their role requires, logged and revocable once a case closes. It is the same rights-managed access model already used for sensitive commercial data, applied to a courtroom.

03What Trusted KnowledgeVerse would add, and why grounding matters here specifically

Once a case's evidence is enriched, Trusted KnowledgeVerse is what makes it conversational. A judge, a clerk or counsel could ask a governed evidence base a direct question, such as what a defendant said about a transaction between two specific years, and receive an answer grounded in the certified record, sources cited, rather than reconstructing that answer from a bundle years thick.

Answers grounded in citations are one layer of protection against hallucination. We are also developing ontological grounding, building knowledge graphs that map how facts, parties, dates and documents in a case relate to each other, rather than treating each document as an isolated block of text. A model reasoning over a proper knowledge graph is constrained by the actual relationships in the evidence, not just by what a retrieval search happens to surface. That structural constraint is what closes the gap a citation alone cannot, it makes fabrication considerably harder to produce in the first place, not just easier to catch afterwards.

The grounding question is not abstract in Singapore. At least two civil cases here have already involved lawyers found to have used generative AI in preparing submissions, with the AI fabricating case citations and legal authorities that do not exist. Both lawyers were sanctioned and ordered to pay personal costs. The Singapore Judiciary's Guide for Using Generative AI in the Legal Sector, published this year, now advises users to cross-check every AI-generated legal reference against reliable sources rather than trusting the tool alone.

An AI tool reasoning over a case's own certified evidence is not guessing at what a document says. It is citing it.

We think the underlying idea extends further than any single case. For a matter of any size, large or small, simple or complex, the evidence, the arguments and the counter-arguments should be able to build up and enrich continuously over the life of a case, rather than sitting as a static bundle assembled once and then frozen. Every stakeholder involved, judge, counsel on both sides, litigants, should be able to hold a natural language conversation with exactly the information they are entitled to see, no more and no less, governed by strong digital rights controls that enforce who can access what at every stage.

Taken further still, good information grounding and a properly mapped set of relationships between facts and documents open up a possibility worth naming plainly. A system built this way could, in principle, help surface submissions that do not hold together, arguments that contradict evidence already on record, or claims that lack the documentary support they assert to have, flagged for a human to examine rather than resolved automatically. That is a meaningfully different proposition from an AI drafting arguments. It is closer to an AI that helps keep everyone honest about what the record actually shows.

04What Trusted ActionVerse would add, and how it extends what already exists

The Singapore Judiciary is not starting this from zero. The Harvey.AI collaboration already produces case summarisation tools for Small Claims Tribunal users and self-represented litigants. Document translation into three languages is already live. These are, in effect, early agentic workflows, narrow automations sitting on top of court data.

Trusted ActionVerse is our no-code layer for building workflows like these, without a bespoke engineering project for each one. Evidence review, summarisation, translation and drafting support within the yellow zone Mr Tan has described, along with administrative workflows around scheduling and disclosure, become buildable on top of a properly enriched evidence base rather than each requiring its own procurement cycle.

None of this replaces judicial judgement. What it replaces is the manual scaffolding currently sitting underneath that judgement, the printing, the physical compiling, hours spent re-reading years of correspondence that a governed AI system could surface in seconds, with the source cited and the judge or lawyer still making every substantive call.

Singapore's courts made the decision to go digital decades ago. An active AI collaboration, live translation tools and a published generative AI guide put the system ahead of most jurisdictions. What remains is the less visible layer, turning digitised documents into governed, conversational knowledge. We've spent 25 years building Evidence Act certified digitisation for other regulated industries, and that layer is where we intend to focus next.