
Extracting legal principles from judgments requires more than asking an AI model to summarise a case. That constraint underpins the Singapore Academy of Law’s (SAL) expansion of LawNet 4.0’s legal research tools beyond case law to natural-language search across legislation and Academy publications, one year after the platform’s launch.
At TechLaw.Fest 2026, Kenta Kusano, CTO of SAL and CEO of LawNet Technology Services, introduced the new capabilities intended to simplify legal research. He said 65% of LawNet subscribers had migrated to LawNet 4.0 and urged remaining LawNet 3 users to migrate before the older platform’s retirement in December.
Expanded search
LawNet’s Q&A Search now covers Academy publications. As with case law, users can ask a question in natural language and receive a response on the left of the screen, with the sources used to generate it displayed on the right.
“We have integrated this with our e-reader so that you can navigate to the chapter in that book that the answer is referencing,” Kusano said.
The legislation module has also been expanded to connect statutory provisions with relevant court judgments and book chapters. When users select a provision, they can see related cases and publications alongside the legislative text.
Demonstrating the enhancements, Michael Lim, Senior Deputy Director (Product Management), LawNet Technology Services, explained that seven pieces of legislation have been added to the platform this year, bringing the module’s coverage to 10.
Using Section 300 of the Penal Code, which concerns murder, Lim showed how selecting a provision displays its text in the middle of the screen and lists judgments citing that provision on the right.
These connections allow practitioners to move between statutory provisions, judgments, and relevant book chapters without repeatedly searching across separate resources.
Annotated statutes
LawNet also introduced its Annotated Statutes feature, which uses AI to extract legal principles from judgments associated with particular statutory provisions.
Michelle Chua, Senior Assistant Director in SAL’s Legal Research Department, said work on the product began in 2024. After seven months of development, the team had not arrived at a legal engineering workflow it was confident in. The team wanted to avoid feeding an entire judgment into an LLM and simply asking it to explain the judgment.
“As lawyers, we know judgments are very long pieces of text, and they contain a lot of information besides legal principles, including facts, parties’ arguments, and issues in dispute. What we really wanted this product to do is just tell you the legal principles relevant to a particular statutory provision,” she explained.
The LawNet team subsequently developed a proprietary algorithm to produce more focused annotations. The Companies Act was selected as the initial legislation because its length and complexity offered a manageable starting point for developing and validating a method that could eventually be extended to other statutes.
During the demonstration, Chua showed how users could navigate to a provision, such as Section 157 on directors’ duties, and browse AI-generated legal principles organised using tags. These tags help practitioners identify material relevant to their research or explore the principles more broadly.
The tool also includes a source-checking function that lets users reveal the original paragraph from which a legal principle was extracted and compare the AI-generated statement against the judgment’s actual wording.
Chua said the language model was instructed to restate the principles with minimal extrapolation and avoid unsupported opinions about the law. The aim is to preserve the substance of the original text while making it easier to identify relevant legal concepts.
Chua also compared the time spent manually reviewing the principles for evaluation with the time the LLM took to generate a comparable number.
“We had two lawyers at SAL go through about 460 sets of these principles manually. We had to do that to evaluate how well our LLMs were doing, and that took us a combined 132 hours. The LLM can produce a comparable number of these statements of principle in two hours,” she said.
Nevertheless, Chua stressed that the tool is not a substitute for a lawyer’s judgment or comprehensive legal research.
“I would say it is quite accurate, but it’s still not 100%,” she noted, emphasising the need to cross-check the generated principles against their original sources.
She also identified a limitation in how the system handles statutory amendments. At present, it does not account for cases associated with a provision’s previous number when that provision has been renumbered. Practitioners must therefore treat the annotations as a starting point for research rather than an exhaustive account of the law.
Chua described the feature as a focused research aid rather than a direct replacement for a traditional annotated statute book, which may offer broader analysis of legal developments, comparisons across jurisdictions, and interpretative insights.
Developing the annotation method
Kusano said the team had tried uploading PDFs of cases to an LLM to generate annotations, but the results were unsatisfactory.
“We tried that, and it didn’t work so well. We went through a process of trial and error, and in the end we had to come up with a proprietary algorithm to generate the annotations,” he recalled.
SAL has filed a provisional patent application for the methodology used to generate the annotations.
“We are here to support the community as a whole by providing tools that help lawyers with their work. Our aim for LawNet is to provide accessible and authoritative legal research for Singapore,” he concluded.









