Robotic justice? England's top judge opens doors to AI-run courts
Sir Geoffrey Vos, Master of the Rolls and head of civil justice in England and Wales, announced in October 2026 that he intends to retire after nearly thirteen years on the civil appeals court — and is leaving behind one of the most profound debates about the future of British courts: the possibility of artificial intelligence systems deciding civil cases. The statement, made on the stage of the Legal Geek conference on October 14, was not a casual gesture. It was the conclusion of months of direct observation, including a visit to the Intermediate People's Court in Shenzhen, China, where courts already use AI platforms to help judges process hundreds of thousands of civil disputes.
The Chinese model in practice
Vos's experience in Shenzhen was no tourist visit. According to the official speech published by the British Judiciary, the judge observed a live demonstration of an AI-powered platform capable of automatically extracting the relevant points from documentary evidence, listing the issues that, in the machine's view, needed to be resolved to settle the case, and generating drafts of the legal reasoning and court orders that the judge might adopt. The result, described in detail in the publication, is that judges no longer flip back and forth between documents and systems to establish the facts of a case — review has become a matter of checking, correcting, and supplementing, turning work that once took a full day into tens of minutes.
This model is not hypothetical: it is already operational in Chinese courts and has been adopted as part of a national strategy for judicial digitalisation. Shenzhen's court platform operates as a single system for all types of civil case, something Vos described as surprisingly viable, given that it had always seemed improbable a single system could cover such a diverse range of litigation categories. The efficiency is striking, but the question that remains — and which Vos does not shy away from — is how far automation can go without compromising the foundational principles of common law.
The conflict between principle and economics
At the heart of Vos's proposal lies what he describes as a fundamental conflict between principle and economics. The principle is straightforward: human judges read the documents in their cases and then decide them. There is no moral shortcut in that process, and justice, in the British model, is inseparable from the human act of listening, weighing, and deciding. But economics — or rather, the fiscal reality of courts — says something different. Caseload backlogs grow without bound, costs spiral upward, and demand for justice rises alongside the proliferation of AI tools among self-represented litigants. According to Vos, the use of reliable machines that save judges time — including in situations where judges do not have time to verify what the machine is telling them — is what could clear backlogs and deliver cheaper justice.
Vos emphasised that the distinction between mechanical tasks, such as transcription, case management, summaries, and chronologies, and tasks that form part of the decision-making process, is easier to articulate than to define. "That distinction is more easily said than actually defined," he wrote in his prepared remarks. And precisely that ambiguity makes the question so difficult: if AI can summarise a case in thirty seconds, filter relevant precedents, and produce a draft decision grounded in legal reasoning, where does one draw the line between assistance and replacement? The British judge's answer is cautious. He insists humans must remain "front and centre" of final decisions affecting people's lives, but admits it would be natural — even desirable — for individuals and businesses to prefer, at least for minor disputes, having their cases decided by machines, once they understood those resolutions would be "far quicker and cheaper than waiting for human judges to decide outcomes using outdated analogue processes and procedures".
The hallucination risk and the trust problem
Vos's argument encounters an immediate and compelling counterpoint: the case of US federal judge Henry Wingate in Mississippi, whose ruling was vacated by the Fifth Circuit after it was discovered that one of his law clerks had used an AI tool — specifically Perplexity — to draft an order. The result was a decision riddled with AI hallucinations, including citations to cases that did not exist and errors of law. Judge Jerry Smith, writing for the appeals panel, spared no words: he called the decision "a botch job" and noted that while the temptation for the judge to want to prove that, even with correct law and fact, his initial view should prevail was understandable, the integrity of the system was at stake. The case was reassigned to a different judge.
This episode is not isolated. A survey conducted by the National Center for State Courts, with interviews of judges across ten American states, revealed that most judges who experimented with using AI for legal research received back a fabricated case as a response. Some reported requiring their law clerks to sign "acceptable use" agreements to mitigate the risk. Every judge interviewed was fully aware of the hallucination risk.
Vos knows this and nonetheless downplays it. In an earlier February 2026 talk, he characterised reports of fictitious authorities generated by large language models as something "pointless to dwell on" — anticipating, as he did, that hallucinations will be "things of the past" very soon. This technological confidence is not unfounded: advances in the accuracy of large-scale models throughout 2025 and 2026 have drastically reduced hallucination rates. But the question is not merely technical; it is institutional. A calculation error is correctable. A judicial error based on a poorly reasoned AI decision can mean the deprivation of fundamental rights.
The future of civil justice
What makes Vos's proposal particularly timely is the context in which it emerges. Less than a year ago, the Master of the Rolls himself warned of a flood of civil claims generated by self-represented litigants who use ChatGPT and Copilot as their first port of call — replacing lawyers — and transform masses of documents into legally arguable claims, all for free. The consequence, Vos predicted, is already visible: an exponential increase in civil, family, and tribunal claims. British courts, with tight budgets and insufficient staff, are under pressure to absorb this demand.
The response Vos proposes is neither utopian nor hasty. He acknowledges it will take time for courts to develop AI programmes that can be calibrated, tested, evaluated, and authorised to assist — let alone replace — judges in their decision-making. The British judge also emphasised that all models currently under discussion begin with a foundational statement: AI can help the judge, but must never replace the judge. And that judges will always have the last word.
The question is whether that dividing line — between assisting and replacing — survives the test of time. When an AI system can process a civil case in minutes and produce a draft decision that, in practice, a human judge merely reviews and signs, the distinction becomes more philosophical than functional. And what if litigants prefer it? If businesses and individuals opt for machine resolutions, faster and cheaper, will the human system remain the default destination or be gradually marginalised to the most complex and contentious matters?
Geoffrey Vos's legacy, who retires on October 31, 2026, will likely be measured not by the judicial career he has ended, but by the question he has left open: when artificial intelligence can decide civil cases as reliably as humans and, certainly, much cheaper and faster, will justice cease to be a human act and become an automated service — or will the court remain, as Vos insists, as the last bastion of human judgment?
What happens to the rule of law when the person who decides your case is not a person at all? That is the question Vos's retirement speech forces every reader — and every policymaker — to confront.
Sources: Judiciary UK, Law Gazette, Bloomberg Law
✓ Independent sources cross-checked and verified before publishing
