Geoffrey Vos uses a newly published WIRED interview to argue for a public debate about AI deciding some disputes. He describes daily Copilot use for administrative work, but says he does not use it to write judgments. His case for machine decisions depends on consent or society choosing that route through legislation. WIRED interview, October 28
That is an argument about where courts might go, not an announcement that a new automated adjudication system has been authorised. Reading it alongside the official material reveals two different questions: how judges can use assistance responsibly, and whether a machine should ever exercise decision-making authority itself.
An efficiency argument with a longer history
The position has a documented history. In an April 30, 2026 speech to the Judicial Institute for Scotland, Vos predicted that economics would push routine judicial work towards machine assistance or direction. He expected dedicated programmes to handle some ordinary disputes more quickly and cheaply, initially with the parties’ agreement. Those were his projections, not independently demonstrated performance results. Official speech
In the same speech, he distinguished final judicial decisions from administrative ones: a court can be a citizen’s last opportunity to assert rights. He argued that surrendering human determination required informed acceptance, debate and legislative change. He also envisaged continuing human roles in appeals, developing legal principles and helping people understand decisions. That older material makes the current discussion a continuation of an explicit institutional argument, rather than the first appearance of the idea.
The important unresolved question is what counts as suitable work. A task being repetitive does not, by itself, establish that its evidence is uncontested or its consequences minor. An efficiency claim needs to identify the particular decision being accelerated before speed becomes a meaningful measure of success.
Existing guidance keeps responsibility with people
The judicial guidance dated October 31, 2025 provides a more immediate operational reference. It permits potentially useful assistance with summaries and administrative work, but requires checking accuracy. Judicial office holders remain personally responsible for material issued in their names and must read the underlying documents; AI cannot substitute for direct engagement with evidence. It also cautions against unverifiable legal research and public chatbots’ legal analysis. Official judicial guidance
The document addresses confidentiality, fabricated authorities and bias, and warns about hidden instructions in material processed by AI. These cautions concern the integrity of work entering the judicial process. They do not amount to a tested framework for handing the final decision to a model. This is dated background, not a fresh October 2026 policy change.
The distinction is substantial. A checked summary supports an accountable person’s review. A binding outcome produced by a machine would require answers about authority, contestability and responsibility that the usefulness of a summary alone cannot supply. Success at one task should not silently become permission for the other.
Consent and evidence are separate tests
In the interview, Vos also identifies family and complex commercial cases as areas where human judgment matters. His optimism is therefore not a claim that every dispute should be automated. Full interview
Even a narrower proposal would need both public legitimacy and evidence about its operation. Agreement to participate would not establish accuracy; a strong accuracy result would not establish that participants understood what they were accepting. A credible evaluation would need to examine both, while making clear how an adverse result could be challenged.
The interview renews that debate without resolving it. The existing documents offer an accountability baseline, while Vos’ proposals ask whether the boundary could move. Any subsequent proposal should be judged on its defined scope, safeguards and demonstrated outcomes, not treated as an inevitable consequence of judges already finding AI useful at their desks.










