An AI judge is no longer a thought experiment in England and Wales. In a wide-ranging WIRED interview published in late October 2026, Sir Geoffrey Vos, the second most senior judge in the country, argued that an AI judge should one day decide straightforward civil claims, from personal injury cases to low-value professional negligence disputes.
Vos retires this week as Master of the Rolls, the most senior position in England’s civil justice system. As one of his final acts, he is trying to force a debate most of the legal profession would rather postpone: whether, and how, judges should be using artificial intelligence, and when an AI judge should be allowed to decide a case.
This guide explains what Vos actually said, what England’s own guidance permits today, why chatbots are already changing what arrives at county courts, where an AI judge line might be drawn, what other countries have tried, and what businesses should do now.
Table of contents
- Who Is Geoffrey Vos, and What Did He Say?
- What the Rules Allow Judges to Use AI For Today
- The Chatbot Flood Building at County Courts
- Deepfakes and Fake Evidence: ‘The Stuff of Judging’
- Where to Draw the AI Judge Line
- Why the Legal Profession Is Right to Be Nervous
- AI Judges Around the World: What Already Exists
- What Businesses Should Do Before AI Judges Arrive
- Our Verdict
- FAQs About AI Judges
- References
Who Is Geoffrey Vos, and What Did He Say?
Geoffrey Vos is the Master of the Rolls, the senior judge responsible for the civil courts of England and Wales. He has spent decades on commercial benches, led the modernisation of court procedures during the pandemic, and has spent the last years of his career talking about technology in the justice system.
His core claim is blunt. “What most people desperately want in civil disputes is resolution,” Vos told WIRED. “The question is, where is the line to be drawn?” He has previously described a future in which an AI judge decides simpler civil cases as inevitable, and he says courts across the globe should seriously consider following the Chinese example.
He is also careful about how change happens. “Lawyers are conservative with a lowercase c,” he said. “But that doesn’t mean that they should be backward-looking or retrogressive. It means they should be adopting change cautiously, with their eyes open.”
The Shenzhen Trip That Sharpened His Argument
Before making the argument publicly, Vos visited a court in Shenzhen, China, where civil disputes are already handled with heavy assistance from AI. Local reporting has said Shenzhen judges using the technology have processed cases around 50 percent faster. Vos was impressed, and he told WIRED he believes courts elsewhere should consider following.
That visit matters because it turns the AI judge debate from philosophy into a comparison. There is no longer a hypothetical about whether machines can speed up dispute resolution. Some of the largest courts in the world are already running the AI judge experiment at scale, with human judges still nominally in charge.
What He Uses AI For Himself
Vos says he uses Microsoft Copilot constantly, but never for the part of the job that decides outcomes. He uses it to locate documents filed years ago, to navigate enormous volumes of material, and to check things over. With eleven meetings a day, he says the time saved is real and he does it daily.
What he will not do is use AI to write judgments. He is also cautious about using it to read the papers for his own cases, because he sits on the most important hearings in the system. For Vos, the boundary is clear: AI may help a judge prepare, but the decision and its reasoning belong to the human, not to a machine that might otherwise sit as an AI judge.
What the Rules Allow Judges to Use AI For Today
England’s judiciary already has written policy on this question. The current Artificial Intelligence Guidance for Judicial Office Holders, published in October 2025, tells judges to confine their AI use to administrative tasks and background work, and treats anything touching the AI judge question with deep caution.
The table below summarises the practical boundary that guidance draws, and where Vos personally sits within it.
| Task | England’s guidance | Vos’s own practice |
|---|---|---|
| Finding filed documents | Allowed as administrative work | Uses Copilot daily |
| Summarising large bodies of text | Allowed with caution and verification | Uses it regularly |
| Reading case papers | Treated with caution | Avoids it for his own big cases |
| Drafting judgments | Not permitted | Does not do it |
| Deciding cases | Human judges only | Wants a public debate about the future |
Why Judges Move Carefully
Vos is explicit about why the caution exists. Public confidence in the justice system rests on the belief that judges have absolute integrity, and that integrity survives only if courts explain how they are adopting new technology and build a broad consensus that it is appropriate.
That is why his intervention is deliberately provocative. A serving judge of his seniority does not float a proposal like the AI judge to make headlines. He is testing whether the profession, the public and Parliament are ready to answer the question he thinks is unavoidable, and whether the consensus can be built before rather than after the first AI judge ruling.
The Chatbot Flood Building at County Courts
The most immediate effect of AI on the courts has nothing to do with robots in robes. It is the volume of claims arriving at county court counters, and Vos gave a striking figure: the number of county court claims started rises around 10 percent every year, and it looks set to increase.
The reason is simple, and it has nothing to do with the AI judge itself. “Everybody who has a legal problem now puts it into Gemini, ChatGPT, or Claude,” Vos said. A chatbot can draft particulars of claim, assemble the case narrative, and produce a claim form. Some of the resulting claims are weak or bogus, and court staff are absorbing the consequences.
Not Every Bot-Born Claim Is a Bad Claim
The instinctive reaction is to treat machine-made claims as time-wasting. Vos pushes back hard. “It’s a mistake to assume that just because a claim is created by AI, it’s totally frivolous,” he said. If a problem is worth bothering ChatGPT about, the person probably genuinely believes they have a claim.
Explore the problem in detail, he notes, and it may turn out to be a different legal issue than the one the chatbot described. But it may still be a real legal problem, and wanting to bring it to a court’s attention may still be reasonable. “People are not generally time wasters,” he said.
There is also a history lesson here. For most of its existence, formal justice has been a service available largely to rich people. Chatbot-assisted claims represent a genuine expansion of access to justice, which is exactly why they strain a system built for the opposite problem of scarcity. That tension is the real engine of the AI judge debate: the same technology that helps claimants reach court also helps courts that cannot staff their counters keep up. Our overview of how the US, China and the EU are taking three different paths to regulating AI shows how differently societies absorb this kind of pressure.
Deepfakes and Fake Evidence: 'The Stuff of Judging'
The other chatbot effect frightening lawyers is fabricated evidence: images, video and voice recordings that never happened. Courts have historically been able to rely on the difficulty of manufacturing convincing evidence, and generative AI has removed that difficulty overnight, which also weakens the case against ever letting an AI judge weigh such material.
Vos’s reaction surprises people. There is “absolutely no difference, in scale or qualitatively,” he argues, in somebody producing a false image made by AI rather than a false photograph from a camera. Judges have always had to work out whether a person is telling the truth, and he says humans are not bad at lying.
Forged Wills, Perjury and Prison
“Forged wills” were the deepfakes of earlier centuries, Vos points out. Courts have spent generations learning to test documents, recordings and witness accounts, and the penalty for manufacturing evidence has never changed. People who produce fake images to a court will find they are committing perjury and will go to prison, just as forgers of wills always did.
The technology is also racing to close the other side of the gap, where a research breakthrough in deepfake detection suggests machines may get better at spotting fabricated video than the human eye, and a cybersecurity expert’s guide to spotting AI-powered scams shows how forgery literacy is spreading beyond the courtroom. Vos’s position is that courts absorb new categories of lies the way they always have: through evidence, cross-examination and consequence, a lesson every AI judge proposal has to survive before it is tried.
Where to Draw the AI Judge Line
If an AI judge is on the table, the whole debate lives or dies on where the line sits. Vos’s own answers are more textured than either fans or critics of the idea expect, and they split along two axes: how much human judgment a case type requires, and whether society has consented to the switch.
The table below maps his position across the main categories of civil dispute he discussed.
| Case type | Vos’s position | Why |
|---|---|---|
| Personal injury claims | Strong candidate for AI adjudication | Formulaic damages, high volume, claimants want resolution fast |
| Low-value professional negligence | Good candidate | Cost of a human trial can dwarf the dispute itself |
| Complex commercial disputes | Human judge expected | Novel points of law, huge stakes, precedent value |
| Custody and family cases | Essentially human | Expected by society, including the younger generation |
| Everything in between | Gradual, consent-led whittling away | AI collates information; humans decide; the boundary moves with consensus |
The chart below renders his position as a spectrum, so you can see the shape of the argument rather than the individual answers.
Consent, Legislation and the Cultural Grain
Even for the best-fit categories, Vos does not think a government should simply flip the switch. “It will only be possible to ask people to accept algorithmic decision-making if they consent to the process in a particular case, or if society decides through legislation,” he said. Delivering justice must sit within the cultural grain, he argues, or the system loses the people it exists to serve.
He also names the psychological risk directly. When people lose a case before a human, even an outcome they regard as wrong can be accepted over time because it was delivered in a human way. An adverse machine-driven decision may be harder to swallow, which is precisely why the AI judge conversation needs consent designed into it from the start.
The Explainability Objection
The most common objection to an AI judge is that nobody can see how a model reached its verdict. Vos’s answer is that the debate should be about outcomes, not hidden reasoning. The AI can produce very plausible reasoning, he concedes you cannot see how it was produced, but you can read the reasoning it produced and judge its plausibility in the open.
Critics will hear an under-answer in that, and they have a real case: plausible reasoning is not the same as accountable reasoning, and an AI judge that cannot be interrogated is harder to appeal than a human who can be. The counter-example they reach for is American. In State v. Loomis, the Wisconsin Supreme Court held in 2016 that a judge could consult a proprietary algorithmic risk score before sentencing, but only with warnings, and never as the sole factor, because the score’s logic could not be examined.
Why the Legal Profession Is Right to Be Nervous
The profession’s anxiety is not abstract. Around the world, courts have caught lawyers citing case law that does not exist, invented by generative models and filed as if it were real. Sanctions have followed, from fines to referrals to professional disciplinary bodies.
In 2026, in what was reported as a first in the UK, a lawyer was struck off in part for misusing AI, including the submission of false authorities generated by the technology. A batch of Mississippi attorneys faced sanctions in June 2026 in a widely reported wave of the same failure. Each incident reinforces the same lesson about the three big types of AI risk experts warn about: capability outrunning competence is a failure mode, and the legal system pays for it in wasted court time.
The Fabrication Pattern That Keeps Courts Awake
The fabricated-citation cases follow a recognisable pattern. A busy practitioner uses a model to draft quickly, the model produces confident citations, nobody checks them against a real database, and a judge or opposing counsel discovers the citations are fiction. The failure is not the model producing plausible nonsense. It is the human filing it without reading it.
That pattern is also the best argument for the careful version of Vos’s proposal. If the concern is competence under pressure, a supervised system that checks every citation against real law may beat an unsupervised human. If the concern is legitimacy, no amount of accuracy settles it, least of all when the proposal on the table is an AI judge with no witness box of its own. The political fight over AI regulation in the US is a reminder that legitimacy fights get settled by politicians, not by benchmark scores.
AI Judges Around the World: What Already Exists
The AI judge is not something the world has never seen, and Vos knows it. He is describing a spectrum that different legal systems have already entered at different points, and England has not chosen its position yet.
China runs specialist internet courts, opened in Hangzhou in 2017 and followed by Beijing and Guangzhou in 2018, where litigation over online disputes happens largely online, with machine assistance in filing, evidence review and procedure. Shenzhen, the court Vos visited, represents the next stage: AI assistance woven through ordinary civil judging.
Estonia, in 2019, went further in one narrow lane. The country’s ministry of justice published plans to develop a robot judge for small claims under 7,000 euros, with the stated aims of speed and consistency. The project stayed experimental, but the template it offered, a capped category with routine facts, is exactly where an AI judge fits first in Vos’s ordering.
State v. Loomis: The Warning Label From America
America’s most influential encounter with algorithmic judging was not a robot judge at all. It was a score. In State v. Loomis, Eric Loomis challenged a sentencing decision informed by COMPAS, a proprietary risk assessment tool. The Wisconsin Supreme Court ruled against him in 2016 but attached strict language: the algorithm could not be the sole factor in a decision, and courts had to be warned that its private logic could not be tested.
Loomis is the case every AI judge proposal has to survive, because it establishes both halves of the doctrine at once: machines already inform real decisions, and courts may only lean on them with disclosure, limits and a human who remains answerable.
What Businesses Should Do Before AI Judges Arrive
Court modernisation moves at the speed of legislation, but contract drafting moves at the speed of a red pen. Businesses that resolve disputes constantly, consumer claims, service credits, routine negligence matters, should assume the AI judge era will reshape the resolution landscape within this decade, and the shift favours whoever prepared.
Our AI strategy work with organisations lands on the same four questions every time, and they are worth putting in front of your dispute counsel this quarter rather than next.
A Practical Checklist
First, map your claims profile: volume, average value, and which disputes are formulaic enough to be adjudicated by rule. Those are the categories an AI judge, or a private AI arbitration service, would absorb first. Second, review dispute resolution clauses for terms that assume human-only adjudication, and for arbitration language that would even permit machine decisions.
Third, build evidence provenance now: signed records, tamper-evident chains, and communication trails that survive the deepfake era. If your evidence cannot prove where it came from, its value in any future courtroom collapses. Fourth, pilot private AI-mediated resolution for low-value supplier and customer disputes, where both parties consent, before any of it is imposed on you. The teams that trial machine justice voluntarily will know which parts of it work.
Our Verdict
Geoffrey Vos is not asking you to love the AI judge. He is asking a narrower question with a much bigger tail: if resolution is what people want from civil justice, and machines already resolve some of it faster and more consistently than exhausted humans with overloaded diaries, on what principled line do we keep refusing?
His own answer, a consent-led, category-by-category whittling away, is the most credible version of machine justice currently on offer. The strongest objections are not about speed or accuracy; they are about legitimacy, explainability and the moment a state’s monopoly on judgment starts quietly transferring from an open courtroom to an AI judge no citizen can audit. The Shenzhen precedent shows the practical path; Loomis shows the legal guardrails.
For now, the AI judge remains exactly what Vos called it: a debate to be had, in the open, internationally, before the facts on the ground have the argument for us. England’s most senior civil judge has now put his card on the table. The rest of us have to play the hand.
FAQs About AI Judges
Would an AI judge replace human judges?
No proposal on the table says that, and Vos explicitly rejects it. He expects human judges permanently, with something almost biblical in the human desire to be judged by a fellow human. The realistic change is narrow: simple, high-volume civil categories moving to an AI judge first, with consent.
Can a judge in England use AI to decide a case today?
No. Current judiciary guidance confines AI use by judicial office holders to administrative and preparatory tasks. Decisions remain human work, and the seniority of the person calling for that boundary to be debated says more than any rule change.
Are lawyers already getting in trouble over AI?
Yes. Courts have sanctioned lawyers for filing AI-hallucinated case citations, and in 2026 a UK lawyer was reported as the first struck off partly for AI misuse, including false authorities. The failure pattern is always the same: unverified machine output filed as if it were truth.
Which cases would go to an AI judge first?
The consensus candidates are personal injury claims and low-value professional negligence in England, plus capped small claims as trialled in Estonia. Custody, family and complex commercial disputes sit at the far end of the line, and the whole boundary is set by consent or legislation, not by capability.
References
Should AI Decide Your Court Case? This Judge Says Yes (WIRED)
Artificial Intelligence Guidance for Judicial Office Holders (Judiciary UK)
Speech by the Master of the Rolls: Artificial Intelligence and the Judiciary (Judiciary UK)
AI helped Shenzhen judges handle cases 50% faster (South China Morning Post)
SDT strikes off lawyer over AI-generated false authorities (Kennedys Law)
Mississippi sanctions lawyers over AI-generated filings (The New York Times)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
