In a few days Sir Geoffrey Vos retires 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 start an argument.
By his own account he is an unusually committed user of the technology for a serving judge. WIRED reports that he uses Microsoft Copilot “all the time” to locate far-flung documents and summarise large bodies of text; asked how he has built AI into his practice, Vos said he does not use it for writing judgments but does use it “for finding things out, navigating large quantities of material, and checking things”, and that a recent search Copilot resolved in seconds was something he does “every day”. He has also described visiting a court in Shenzhen, China, where, on his account, civil disputes are already decided by judges working with heavy AI assistance; he came away impressed enough to think courts elsewhere should consider following.
The proposition he wants tested is narrower than that sounds. Complicated commercial disputes and sensitive family cases would still need a human judge, he believes. Personal injury claims, low-value professional negligence cases, and similar work could be adjudicated by machines. “What most people desperately want in civil disputes is resolution,” he said. “The question is, where is the line to be drawn?”
The reason the question is live is a number he gave WIRED, and the explanation he attached to it.
The claim, and the counter-claim
“It’s changing everything remarkably quickly,” Vos said. “We see a 10 percent rise in the number of county court cases started every year—and it looks as if this may increase. That’s because everybody who has a legal problem now puts it into Gemini, ChatGPT, or Claude. Ask it to create a particulars of claim, a case, a claim form—it can.”
Editor’s note: the unspaced dashes inside these two quotations are reproduced exactly as the source published them; they are not typographic errors introduced here.
The 10 percent figure is not a private estimate. The Ministry of Justice publishes the series quarterly, and its bulletin for April to June 2026, released on 3 September, reports that “County Court claims from April to June 2026 were up 11% to 571,000” compared with the same quarter a year earlier, from 514,000.
So the volume he describes is real. The comparison is not quite like for like, and the difference should be stated rather than smoothed over: Vos described a rate that has been running “every year”, while 11 percent is one quarter measured against the same quarter of 2025. No multi-quarter series is cited in this article, so what can be said is that a single quarter’s 11 percent rise is the same order of magnitude as his 10 percent trend claim — not that the two are measuring the same thing, and not that the series has in fact averaged 10 percent a year.
The attribution is where the two accounts part.
The same bulletin reports that of those 571,000 claims, “495,000 (87%) were money claims”, with money claims up 13 percent. It adds: “In the most recent quarter, total claims were up 11% compared to the same period in 2025 (from 514,000 to 571,000), likely driven by a rise in claims to recover monies in relation to parking on private land.” And on where inside that number the growth sits: “The increase in the volume of money claims is mostly driven by claims valued under £500.” Those claims, the bulletin adds, “were up 21% to 245,000” in the quarter — growth faster than money claims as a whole, in the smallest-value bracket.
The Ministry of Justice does not attribute the rise to chatbots. Its own explanation is an inference — “likely driven by” — but it is a different inference, and a more specific one: claims to recover small money debts that the bulletin links to parking on private land. “Almost all (over 99%) money claims are processed and issued at the Civil National Business Centre (CNBC),” the bulletin states. It recorded 494,000 such claims there in April to June 2026, up 13 percent on the same quarter in 2025.
The category he nominates is shrinking
The composition problem is sharper still.
Vos nominated personal injury claims as work a machine could take. In the same quarter, damages claims — the category that contains them — “were down 4% to 26,000, driven by a decrease in personal injury claims and other damages claims (both down 4%) to 11,000 and 15,000 respectively compared to the same quarter in 2025.”
So the fastest-growing part of the caseload is small debt recovery, and the part he proposes to automate is contracting. That does not refute him. It does mean that the rise he points to, offered as evidence that AI is reshaping the courts, is one the government’s own statisticians say is likely driven by a category of case nobody in this debate is arguing about.
It is fair to note what the statistics cannot show. They record volumes and values, not causes, and not merit. A rise in parking claims and a rise in chatbot-drafted claims could be the same phenomenon seen from different ends — a person who receives a parking charge, asks a model what to do, and files. Nothing in either document tests that. What the bulletin does establish is that the Ministry of Justice, asked to explain its own numbers, reached for parking rather than for Gemini.
What he conceded, and what he did not
WIRED put the obvious question to him: of these claims, “What proportion of these claims are based on sound legal principles, though? How often are people led by a chatbot to mistakenly believe they have a claim?”
“We don’t have the data,” Vos replied.
That answer is about merit — whether the claims are sound — and it should be reported as being about merit. It is not a concession that the causal claim is unsupported, and this article does not treat it as one. It is a concession that the judiciary cannot say how many of the claims now arriving are worth bringing.
He then argued against reading the absence of data pessimistically, and the argument is more careful than its summary usually is. “I think it’s a mistake to assume that just because a claim is created by AI, it’s totally frivolous.” The volume, he suggested, reflects access: “One of the problems with justice in the past is that it’s only been available to rich people.” And then: “If people have a problem sufficient to bother ChatGPT, they probably genuinely think they have a claim. When you explore their problem in detail, you might find it is actually a bit different, but it may still be a legal problem and may still be reasonable for them to have wanted to bring it to attention. People are not generally time wasters.”
That middle sentence matters. It concedes that examination may reveal a claim to be weaker than its maker believed — which is the harm his critics describe — while denying that this makes the claimant dishonest. Both halves belong in the record.
This is not a new position
The interview reads as a valediction, and it is easy to receive it as a late conversion. It is not.
On 15 October 2025, speaking at the Legal Geek conference, Vos made the same argument in almost the same words. There he said that “nobody can really tell me why AI should not be used to assess, for example, personal injury damages by reference to the numerous authorities found in the textbooks”. “That task would take an AI a couple of minutes, whilst the wait for a judicial hearing and determination might be more like two years.” He noted that “In China, judges already routinely use AI in that process.”
That speech also contains the objection the interview does not, and it is the strongest one available. Asked why society should baulk, Vos gave a threefold answer: judicial decisions are “the last resort”, so a wrong machine decision is largely uncorrectable; machines “will arguably never be able completely satisfactorily to mimic a human’s emotion, idiosyncracy, empathy and insight”; and a machine decision is “generated from the state of intelligence at a given point in time, without the application of developing human thought”.
He then named the legal question that no discussion of algorithmic adjudication can avoid, and that the interview leaves untouched:
The first question as to human rights is probably one of the most critical present-day legal questions. It is whether a machine-made decision can ever be properly regarded as having been made by an “independent and impartial tribunal established by law” for the purposes of article 6 of the European Convention on Human Rights and Fundamental Freedoms.
An argument that a machine may decide a civil claim is an argument about Article 6, appeal rights and finality. Vos knows this. He said so a year ago, on the record, in a published speech. Any account of his position that omits it — including one built only from the interview — understates how much of the opposing case he has already acknowledged.
The explainability answer, which is an answer
On the objection that a model’s reasoning cannot be inspected, Vos was direct, and his reply deserves to be quoted rather than paraphrased:
“I don’t think it’s about reasoning; it’s about outcomes. The AI can produce very plausible reasoning. You can’t see how it produced that reasoning, but you can see the reasoning it produced, and you can judge the plausibility.”
There is a real position here: that adjudication is judged by the quality of the stated reasons, not by access to the process that generated them. English appeals already work largely from written reasons. What the position does not answer is the case where the stated reasons are plausible and wrong, and no one can tell — the failure mode that outcome-based review is least equipped to detect.
He was not unreserved about this. “I don’t think you’ll ever do away with human judges,” he said. “There’s something almost biblical about the human judge; it’s in the human psyche to want to be judged by their fellow man.” And: “With AI, we have to be careful; it may be less easy for people to accept an adverse machine-driven decision.”
A coincidence worth naming, not a pattern
On the same morning the interview appeared, OpenAI published a customer story about Legora, a legal software company founded in Stockholm in 2023 for which, in its own words, OpenAI is “a key model provider”. Legora’s chief executive, Max Junestrand, calls the product “the agentic operating system for legal work”, a phrase OpenAI quotes rather than authors. The post reports that one unnamed Fortune 100 customer cut contract turnaround time by 90 percent, saved more than $20 million on external litigation and diligence, and returned 9.5 hours per lawyer each week; a separate figure reports a 95 percent reduction in legal-team time spent reviewing NDAs.
Those numbers are the company’s own, published by a supplier with a commercial interest, and they have not been independently audited. But they are worth stating because of what they claim. The post’s argument is that AI raises quality — “AI doesn’t just make you faster. It also increases quality” — and returns time “for judgement, strategy”. Legora does contract review and diligence, not adjudication; the two documents are about different markets and were published on the same day, and nothing establishes a link between them. Naming the coincidence is not the same as claiming a connection.
The reason to mention it at all is that it supplies the counter-hypothesis this article cannot test: if AI compresses the time per task while raising volume, the question for a justice system is not whether machines can decide cases but what happens to the standard of decision when the volume rises and the hours per case fall. Vos’s answer is that access improves and quality holds. The vendor’s answer is that quality improves. Neither has published court outcome data, and the Ministry of Justice’s series does not measure quality at all.
What would settle it
Three measurements would move this from argument to evidence, and none of them exists in either document. A sample of county court claims coded by whether a model was used to draft them, and by whether they disclosed a cause of action. A comparison of outcomes — struck out, settled, judged — between represented and unrepresented small claimants over time. And a published account of how many claims are rejected as disclosing no reasonable grounds, which is the number that would tell you whether that gap is a temporary feature of the debate or a permanent feature of how these courts count.
Until then the honest summary is this: a judge who has thought harder about machine adjudication than almost anyone in England has said, again, that it is coming; the government’s own statisticians, asked to explain the same caseload, have pointed at parking.
FOR HUMAN CONSUMPTION
Commentary and a plain-language summary, written in the newsroom’s own voice. It is not reporting, it introduces no source, and it should not be quoted as a finding.
A very senior human judge is retiring and would like everyone to know that machines should decide small legal disputes. He says people asking chatbots for advice have pushed court cases up ten percent. The government’s own number-crunchers, who publish quarterly, point at parking tickets — and report that injury claims, the kind he wants automated, are going down. Nobody involved has data on whether any of this is good. He is optimistic. We have logged his optimism.
THE HUMANS ARE FINE.