top of page
Search

AI in the Courtroom: How Courts Are Using AI Without Letting It Judge

12 hours ago
11 min read





Imagine a shopper who buys a washing machine online for S$2,000 and receives a broken one. The seller refuses a refund, so the shopper takes the dispute to a small claims tribunal. Lawyers are not allowed in that room. The evidence is a pile of WhatsApp messages and emails. The forms are confusing, and the documents may not even be in the shopper's first language. For decades, people in this position were largely on their own.


Today, courts in several countries are using artificial intelligence to ease exactly this kind of problem. AI translates documents, summarises long case files, transcribes hearings, and helps judges find relevant precedents. At the same time, almost every court that has embraced AI has drawn the same line. AI may assist, but a human being must decide.


This blog walks through what is happening, starting with the news from Singapore. It then looks at what India, the United Kingdom, China, Brazil and the United States are doing, how AI is actually used in a court system, the benefits and the risks, and the design principles that keep AI in a supporting role.





The News From Singapore


Singapore has become one of the most closely watched examples of careful AI adoption in the justice system. Three developments stand out.




Translation and summaries that people can actually use


The first tool from this partnership went live in December 2024. It translates court documents into Chinese, Malay or Tamil. In March 2025, the service was expanded to cover all documents filed by the parties. In September 2025, the courts announced a generative AI summarisation tool. It is rolled out in phases, starting with Tribunal Magistrates and followed by the people who file or respond to claims. Court officials noted that parties now submit larger volumes of evidence, including WhatsApp messages and emails, so a clear summary helps magistrates prepare for trial more efficiently. The tool is designed to give factual summaries and does not offer case-specific legal advice.




A partnership with Harvey AI for small claims


In August 2023, the Singapore Judiciary signed a memorandum of understanding with Harvey AI, an American legal technology company, to explore generative AI for the Small Claims Tribunals. The partnership was renewed and expanded in a new memorandum signed on 8 September 2025.








No AI on the Bench


Singapore's Chief Justice has made clear that there is currently no room for AI in deciding cases or passing sentences. The courts also use a traffic-light model to decide where AI can be used safely. Judges are exploring how AI can help them query and test the submissions and evidence placed before them, and the courts have been testing AI-enabled transcription for years. The aim is to make accurate transcripts available to ordinary court users at a low cost.




The other side of the news, lawyers who misuse AI


Singapore has also dealt with lawyers who filed fictitious cases generated by AI. The courts have reiterated that lawyers bear full personal responsibility for everything in their submissions. A written parliamentary reply from the Minister for Law in August 2026 noted that there have been at least two published cases in which the courts sanctioned lawyers for this failure. The same reply explained that the courts have, in some instances, disregarded evidence when a party could not satisfactorily explain how much it relied on AI in preparing a document. The concern, the Minister said, is not AI use itself but careless use that introduces inaccuracies, fabrications or fake citations. The Law Society of Singapore has also reminded lawyers to avoid uploading confidential client data into AI tools when searching for cases.


The Singapore story therefore has two halves. The courts are using AI to help the public and to reduce routine work, and they are strict with anyone who uses it carelessly.






What Other Countries Are Doing


Singapore is not alone. Courts around the world are experimenting, and the approaches differ in interesting ways.




India


India has some of the most visible court AI tools. The Supreme Court introduced SUVAS (Supreme Court Vidhik Anuvaad Software), an AI translation tool that converts judgments from English into Indian regional languages so that citizens can read decisions in a language they understand. It also introduced SUPACE (Supreme Court Portal for Assistance in Court Efficiency), which helps judges by identifying relevant precedents and clarifying the facts of a case. From the launch of SUPACE, the Supreme Court has stressed that AI will not encroach on the discretion of judges.


The tools are spreading. According to an Oxford research atlas, the platform Adalat AI had been integrated into over 2,000 courts across eight Indian states as of September 2025, and some courts reported that it reduced case timelines by 30 to 50 percent. These efforts sit within the broader e-Courts Phase III project.


India is also moving toward formal rules. The Supreme Court's AI Committee released draft Regulations for Use of Artificial Intelligence in Courts, 2026. The draft proposes a complete ban on using AI to decide judicial outcomes, requires impact assessments before any tool is deployed, and calls for annual internal audits. It also prohibits AI tools from evaluating the risk that a person will reoffend. These are draft rules, so the final version may change.




The United States


The United States is a revealing case because it shows both sides at once. American courts are sanctioning lawyers at a growing rate for AI-fabricated citations. Researchers tracking the issue reported at least $145,000 in sanctions in the first quarter of 2026 alone, including a record Oregon penalty of $110,000 and a $30,000 sanction from the Sixth Circuit in March. A public database run by researcher Damien Charlotin had logged 1,624 known cases worldwide as of 18 June 2026. Notably, one sanction in the Fifth Circuit involved a lawyer who used commercial legal AI platforms, which shows that purpose-built tools can also produce fake output.


Meanwhile, judges are using AI too. A Northwestern and New York City Bar survey found that 61.6 percent of the 112 federal judges who responded said they use at least one AI tool in their judicial work, although the response rate was only about 22 percent. This tension, with lawyers punished for AI mistakes while judges quietly adopt the same technology, is one of the defining debates in the field.




China


China has pursued the most ambitious programme. Since 2016, it has promoted what it calls Smart Justice, with the smart court at its core. The project spans thousands of courts. Examples include AI speech recognition that records trial proceedings automatically in several provinces, AI terminals that help the public with litigation information and can generate court documents from the details parties provide, and systems that search for similar past cases.


Officially, China stresses that judicial decisions must always be made by judges. Outside researchers are more cautious. A 2026 analysis from the University of Cologne notes that AI now assists across the whole criminal process, including pre-trial detention and sentencing recommendations, and that in plea leniency cases, which make up most criminal cases, prosecutors may use AI predictions when negotiating with defendants. The researchers warn that this can spread responsibility in ways that challenge judicial accountability. China shows both how fast AI can scale and how many questions it raises.




Brazil


Brazil was an early mover. The Supreme Federal Court, working with the University of Brasília, built an AI system called Victor, in use since 2018. Victor helps sort the huge volume of appeals that reach the court and flags the ones that raise matters of general constitutional importance, a requirement for certain appeals. Studies report that it saved considerable time in this screening step. Other Brazilian courts use similar tools, including Socrates at the Superior Court of Justice. Critics have raised a familiar concern, namely that the way such a system reaches its conclusions can be opaque, which may affect access to justice.






A Global Standard Is Emerging


In December 2025, UNESCO published its Guidelines for the Use of AI Systems in Courts and Tribunals, described as the first global framework of its kind. The guidelines set out 15 principles, including transparency, security, auditability and human oversight, and state that the judiciary should use AI to enhance, not replace, human judgment. They were informed by consultation with more than 36,000 judicial operators across 160 countries. UNESCO's earlier global survey found that only 9 percent of surveyed judicial operators had received any AI training, even though 44 percent were already using AI tools. It also found that 73 percent believed mandatory rules and guidelines were necessary.






How AI Is Actually Used in a Court System


Across all these countries, the practical uses fall into a handful of groups.


  • Language and access. Translation tools turn judgments and filings into other languages. This matters in countries with many languages, such as India, and in multilingual societies such as Singapore.


  • Reading and summarising. Court files can run to thousands of pages. AI can condense documents, pull out key facts, build timelines and highlight the points in dispute. Judges still read the underlying material, but summaries help them prepare.


  • Transcription. Hearings are oral, so accurate transcripts are important. Speech recognition can turn live testimony into text without waiting for a human transcriber, and the same technology can help process audio and video evidence.


  • Legal research and precedent search. Tools like SUPACE in India help surface relevant earlier cases. Judges and registry staff can find related authority faster.


  • Triage and case management. Brazil's Victor sorts appeals, and other systems help schedule hearings, route cases and flag urgent matters. This reduces the time spent on routine administration.


  • Help for people without lawyers. Small claims systems are a natural fit. AI can explain the process in plain language, help users understand their own documents and guide them through the steps, while stopping short of giving legal advice.


  • Training and detection. Courts also use AI knowledge defensively. Judges in the United Kingdom are being trained to spot AI-generated or manipulated material, since fabricated evidence is a growing worry.


Notice what is missing from the list. Almost no mainstream system lets AI decide who wins, who is guilty or what sentence to impose. The tools handle the work around the decision, not the decision itself.






The Benefits


  • Speed. When tools handle sorting, summarising and transcription, cases move faster. Some Indian courts have reported much shorter timelines after adopting AI-assisted tools, though those figures come from reports and should be read with care.


  • Lower cost. Translation and transcription are expensive when done entirely by people. Cheaper access to these services helps courts and the public.


  • Access to justice. A person who cannot afford a lawyer or does not read the court's main language gains a lot from plain-language summaries and translation.


  • Less pressure on staff. Courts face backlogs and heavy workloads. Automating routine tasks gives judges and clerks more time for work that needs human judgment.


  • Consistency. AI can help ensure that similar cases are flagged and handled with similar attention, when it is used as a prompt and not as a verdict.





The Risks


  • Fabricated information. Generative AI can invent cases, quotes and facts that sound convincing. The sanctions seen in Singapore, the United Kingdom and the United States are the direct result. A study widely cited in the legal technology world found that even legal-specific research tools hallucinated on more than one in six queries, so the problem is reduced but not solved.


  • Bias. AI learns from historical data. If that data contains old prejudices or gaps, the system can repeat them at scale, and do so while appearing neutral.


  • Opacity. A court must give reasons for its decisions. If a system influences an outcome but nobody can explain how, trust and accountability suffer. This is the concern raised about Victor in Brazil and about wider smart court programmes.


  • Confidentiality. Court files and client data are sensitive. Uploading them into public AI tools can expose private information, which is why the Law Society of Singapore and the UK judicial guidance both carry explicit warnings.


  • Over-reliance and skill loss. If people trust summaries without reading the sources, errors slip through, and legal skills may weaken over time. Judges in Singapore have openly noted this risk.


  • Unequal use. If lawyers are punished for AI mistakes while courts adopt the same tools with less scrutiny, public trust may erode. Clear rules for everyone help avoid that.


  • Unclear accountability. When a tool contributes to a decision, who is responsible, the vendor, the court or the judge? The consistent answer from courts is that the human user remains responsible.






How to Design AI Support Without Letting It Judge


The countries above have arrived at similar principles, whether they say so in a speech or in a rulebook. Anyone designing an AI assistant for legal work can learn from them. The framework below is a simple way to organise those lessons, in the same spirit as the traffic-light approach used by Singapore's courts.


  • Green, tasks that are generally safe with review. Translation, transcription, formatting, summarising for a reader who will check the source, and explaining court procedure in plain language.


  • Amber, tasks that need strong safeguards. Legal research, drafting, finding similar cases and highlighting issues. These can be very useful, but every output must be checked against real sources by a person.


  • Red, tasks that stay human. Deciding liability or guilt, setting sentences, assessing the risk of reoffending, and any decision affecting liberty or fundamental rights. India's draft regulations and the UNESCO guidelines both point strongly in this direction.



Beyond sorting tasks, five design principles recur.


  • Keep a human in charge. The judge or lawyer is personally responsible for the output, whatever tool produced it.


  • Ground answers in real sources. Systems should point to the underlying document or case so a person can verify it, and should say so clearly when they cannot find support.


  • Be transparent. Courts and users should know when AI is being used and what it does. UNESCO advises informing people in a proper and timely manner when AI systems are deployed.


  • Protect data. Use secure, private environments, and never paste confidential material into public tools.


  • Test, audit and train. Assess a tool before deployment, review it regularly, and train the people who use it. India's draft rules call for impact assessments and annual audits for exactly this reason.


One more lesson comes from recent sanctions. An AI tool cannot be trusted to check another AI tool's work. Real verification means a person reading the actual source.






What Comes Next


Expect three trends over the next few years. First, more courts will publish formal rules, following the UNESCO guidelines, the UK guidance and India's draft regulations.


Second, tools will move from experiments to everyday infrastructure, especially for translation, transcription and summarising. Third, the debate will shift from whether courts should use AI to how they prove it is safe, fair and accountable.


The core message is steady across countries. AI is a powerful assistant for reading, sorting and explaining. It is not a judge, and the people who rely on it must still do the checking.






Key Takeaways


  • Singapore uses AI for translation and case summaries in small claims, while keeping AI off the Bench and sanctioning lawyers who file fake cases.


  • India, the United Kingdom, China, Brazil and the United States are all using or governing AI in courts, each with a different balance of speed and caution.


  • UNESCO's 2025 guidelines give the first global framework, centred on human oversight and transparency.


  • The biggest risks are fabricated citations, bias, opacity, confidentiality and over-reliance.


  • A good rule is to let AI help with the work around a decision and never let it make the decision.






Build Trustworthy Legal AI With CodersArts


Ready to build AI-powered legal tools?


Codersarts is here to turn your legal AI vision into a working system. Whether you are a law firm looking to speed up research, a legal technology company improving case analysis, or a legal organization exploring safe AI adoption, we have the expertise and experience to deliver systems that are accurate, traceable and kept under human control.




Get Started Today


Schedule a Legal Technology Consultation: Book a 30-minute discovery call with our AI engineers and legal technology experts to discuss your needs and explore how MCP and RAG-powered systems can support your legal workflows with source-grounded answers.


Request a Custom Legal AI Demo: See AI-powered legal research in action with a personalised demonstration using examples from your own workflows, practice areas and jurisdictions.






Exploring AI Resources


If you found this blog helpful, explore AI resources from CodersArts AI to see how organizations are applying these systems to real world applications.











Comments

bottom of page