TL;DR

Sir Geoffrey Vos, Master of the Rolls, expects that people will end up accepting machine resolution of low-value disputes for straightforward cost reasons. He wants the profession to settle urgently where the boundary falls between decisions a machine makes and decisions it merely informs, and acknowledged some judges are probably already using large language models beyond what official guidance permits.

The line nobody has drawn

Vos framed the distinction between machine-made and machine-assisted judgments as genuinely difficult, and his reasoning is more uncomfortable than the phrasing suggests. Judiciaries in many countries have issued guidance insisting the judge remains personally answerable for every decision whether or not a model helped. His point is that the guidance describes an aspiration, not a practice.

The mechanism he described is workload rather than misconduct. Research tools of the Harvey and Legora type can answer the question at the heart of a dispute, not merely supply the law around it. Faced with a thousand decisions and finite hours, he argued, the pressure runs one way: toward approving what the machine proposed. AI improves faster than judges can verify what it tells them.

Precedent already exists

The example he reached for is commercial rather than judicial. Last December the American Arbitration Association’s dispute-resolution arm announced a document-only AI arbitrator for two-party construction claims below a set value, with a human reviewing the draft award and testing its reasoning. Vos saw no objection in principle to commercial parties agreeing to that, provided consent is real.

That distinction — freely given consent versus economic necessity — is where his warning bites for ordinary claimants. Consenting to an AI arbitrator as a sophisticated commercial party is not the same as being routed to one because it is the only affordable path to a hearing.

Looking forward

Vos retires this autumn, which makes this something closer to a parting instruction than a consultation. Work is under way at SIFoCC (Standing International Forum of Commercial Courts) to agree principles on what counts as acceptable, and he told the profession to prepare itself urgently.

His closing caution was aimed inward: a legal sector that resists capable AI will look protectionist and self-serving, and public confidence will suffer for it. Coming alongside the Solicitors Regulation Authority’s AI-misuse investigations this week, UK legal services are now being pushed from both directions at once.