Notwithstanding that AI is in its infancy, courts are already dabbling in the use of AI. China’s “Smart Court” programme in Shenzhen reportedly helps judges handle up to 50 per cent more cases each year. Introduced in 2024, it presently supports 85 procedures across civil, administrative and criminal cases, including filing, review and document preparation. In Germany, IBM’s OLGA tool was piloted in 2022 to focus on diesel emissions appeals. It categorises case files, extracts metadata and compares them with earlier decisions, helping judges work more efficiently. In both systems, human judges retain final authority.

Singapore’s courts appear to be more careful and methodical in the adoption of AI. The Small Claims Tribunals began using generative AI in December 2024 to translate notices into Chinese, Malay and Tamil. By March 2025, the use expanded to case documents, including witness statements and evidence. An AI summarisation tool was introduced in September 2025 to assist in the preparation for hearings. By December, AI tools were made available to self-represented litigants.

To govern these tools, the Singapore courts have employed a “traffic-light” model to determine where AI could be safely used. Green for purely assistive use such as translation and summarisation. Amber where closer human oversight is required including use by self-represented litigants to structure their arguments. Red to ensure that judicial decision-making and predictions of case outcomes are made only by judges in person. Singapore’s Chief Justice has made clear that there is presently no room for AI on the Bench, whether in deciding court cases or passing sentence.

Within those green-light boundaries, though, AI’s reach is already considerable. It could democratise and demystify the law by turning dense legal language into plain English, and vice versa, so that people without lawyers can better represent themselves. It could engage directly with self-represented litigants and decide cases without the assistance of legal counsel. Access to justice could then become less dependent on a person’s ability to afford legal representation.

The Question Begged

The issue of AI’s possible future role in deciding court proceedings is one which is faced by all countries and Governments and international organisations are exploring the establishment of standards. In December 2025, UNESCO published the first global ethical and operational framework for AI in courts and tribunals.

All of this points to AI’s presence in judicial proceedings as now settled — what remains is how far it should go. Specifically, whether agentic AI capable of acting independently should be permitted to move beyond assistance as a tool to replace the human judge.

The Case for AI as a Judge

There are obvious benefits to permitting AI to sit in judgment independent of human intervention. An AI judge could operate quickly and without the bottlenecks caused by the limited availability of human decision-makers. It could process many cases simultaneously, retrieve and compare legal authorities at speed, and work without fatigue. It may also command a broader, deeper and more readily accessible knowledge of the law. There would be more consistency in judgments as well.

Importantly, an AI judge is not susceptible to personal bias and emotions entering into judgments.

The Case Against

There are serious risks to unrestrained adoption. The most immediate vulnerability is data dependency. If historical data and legal datasets contain errors, systemic imbalances or outdated biases, be it societal or cultural, AI is likely to codify and reproduce those imperfections at scale.

There is then the question of control. The people or organisations empowered to define or code the legal algorithms and parameters wield immense judicial power. If the code is kept secret, it is incongruent with a foundational principle of justice i.e. decisions must be seen to be made, not simply accepted on trust. This invites the question: who watches the coder?

The distinction between AI as an aid and AI as a decision-maker matters. AI that summarises a case, flags a concern or suggests a precedent to a human judge is fundamentally different from AI that independently makes a binding judgement that affects human liberties. When AI acts alone, it removes the human check that ensures what the algorithm produces is what the law and justice requires. It is this missing check, more than AI’s capability, that should concern us most.

Who Watches Judge AI?

Then there is the question of appeals present in every judicial system. What happens when the algorithm errs? In a conventional court, if a decision is challenged, a different human judge or set of judges consider the appeal and apply their minds to it. An AI appeal may offer no such guarantee as it is more than likely to apply the same algorithm, data or reasoning process. A system that is flawed by design may simply loop back on itself and reach the same decision resulting in the injustice being perpetuated.

There is a further, quieter risk. The human administrator of the AI system could exercise enormous influence over legal outcomes without the visibility, independence or accountability required of judges.

A Cautious Middle Path

Perhaps the answer lies somewhere in between, if not exactly in the middle.

AI could decide high-volume, commonplace, simple, straightforward, lower-value and largely data-driven disputes, such as small claims, routine consumer cases and straightforward minor motor accident claims. However, an accessible right of appeal to a human judge must remain entrenched.

Cases involving fundamental rights, personal liberty or major legal entitlements should remain in human hands. Judging at that level demands discernment, mercy, compassion and conscience, qualities that a reductive system cannot supply. In these instances, AI should serve as the ultimate digital clerk, leaving a human judge to deliver the final decision and answer for it.

The employment of AI in the judiciary requires consideration of a myriad of factors — not least independence, the quality of decision-making, values and fairness. This is not a simple yes-or-no question. It demands rigorous debate. Otherwise, any later attempt to change the law may well be overruled by Judge AI.