Can AI assist judges without deciding cases?
England and Wales' judiciary says approved AI may support tasks such as summarisation, transcription, redaction and document analysis, but every output used in judicial work must be reviewed and approved by the responsible judge. The position sets governance boundaries; it does not publish pilot results or authorise autonomous decisions.
Editorial responsibility: The Impact of AI Editorial Desk · Report a factual concern
At a glance
- 1The position states that judicial decisions must be taken by a judicial office holder and that AI must not displace reasoning, evaluation of evidence or discretion.
- 2Every AI output used in judicial work must be reviewed, verified and approved by the responsible judge, who remains accountable for it.
- 3Approved tools must provide UK data residency, enterprise data protection and disabled web access; new pilots and use cases must pass through a central steering group chaired by the Lead AI Judge.
Research topic
How the judiciary of England and Wales intends to govern AI tools used by judges and judicial bodies, and which decisions remain exclusively human
The answer: assistance is permitted, but the decision cannot be delegated
England and Wales' judiciary has drawn a clear formal boundary: artificial intelligence may support judicial work, but it may not determine the outcome of a case or replace a judge's reasoning, evaluation of evidence or exercise of discretion. The position statement, published on 9 October, says every AI output used in judicial work must be reviewed, verified and approved by the responsible judicial office holder. Accountability stays with that person even when a tool helped prepare the material.
That distinction matters because a summarisation or document-analysis tool can influence what a decision-maker notices long before a judgment is written. Human approval is therefore necessary, but it is not automatically sufficient. A judge can verify only what the interface exposes and what time and training allow. The statement establishes responsibility and a route for oversight; it does not demonstrate that a particular tool is accurate, unbiased or safe in a live case.[1]
The Impact Brief · Free
Follow the evidence in government & policy.
Get a five-minute weekday briefing on what changed, why it matters and where the evidence comes from. Choose the topics you care about.
Which uses have been tested
The judiciary says its work to date has concentrated on lower-risk functions in controlled settings. The listed pilot areas are summarisation, transcription, redaction, writing in accessible language, administrative and leadership tasks, chronology-building and document analysis. The pilots also considered reliability, the amount of human checking required, handling of sensitive material and safeguards needed for operational use. These are support functions rather than automated adjudication, but several can still affect the record a judge sees.
The statement does not identify the products tested, number of users, case types, evaluation design, error rates or subgroup results. It does not say whether outputs were compared with human work, how redaction failures were measured or whether people affected by a pilot could challenge an error. Readers should therefore treat the document as a governance position informed by pilots, not as published evidence that those pilots succeeded.[1]
Approved tools have three explicit technical conditions
Only approved AI tools may be used for judicial purposes. The statement says their features must include UK data residency, enterprise data protection and disabled web access. Those requirements address obvious confidentiality and data-flow risks: case material should not be routed through an uncontrolled consumer service, used casually outside an enterprise arrangement or combined with live web material that cannot be traced. The judiciary deliberately does not publish a product list because tools and versions change quickly.
The three conditions are a floor, not a complete safety case. UK hosting does not by itself establish who can access data, how long prompts and outputs are retained, whether a supplier trains on them, or how model updates are tested. Disabled web access reduces one route for leakage and unreliable retrieval, but an offline model can still hallucinate, omit decisive facts or reproduce bias from its training data. Procurement teams will need more specific controls for each use and information class.[1]
A central group is the gatekeeper for new uses
A central steering group chaired by the Lead AI Judge is described as the sole route for overseeing development, piloting and deployment. It must assess legal, ethical, operational, security and data-protection issues before a new tool, product or use case proceeds. Senior judicial leaders are to assign responsibility for pilots and approved tools, and new uses must be brought to judicial leadership rather than spreading informally from one court or judge to another.
Centralisation can stop a promising demonstration from becoming an unreviewed operational dependency. It can also create a consistent record of versions, incidents and decisions. The statement does not specify membership beyond the chair, approval criteria, publication duties, appeal routes or how often deployed systems will be retested. Those details will determine whether oversight is an active assurance process or simply an administrative checkpoint.[1]
What this means for judges, lawyers and court users
For judges, the practical obligation is to remain able to explain and defend the work as their own. An AI-produced chronology or summary should be checked against the underlying documents, with material omissions and uncertainty visible. Approval should not become a final click after reading a fluent output. Training from the Judicial College will need to cover when not to use a tool, how to test citations and redactions, and how to record the role AI played when that is relevant to fairness or open justice.
For lawyers and court users, the statement offers assurance that a model is not authorised to decide a case. It does not create a right to inspect prompts, model outputs or evaluation records, nor does it explain when use will be disclosed. If a support tool omits evidence, exposes confidential material or introduces a misleading summary, the affected person needs a clear way to raise the issue and obtain correction. That procedural layer is not set out here.[1]
The policy is narrower than guidance about AI-generated filings
The document governs AI tools used by or on behalf of the judiciary and guides senior judicial leaders working with the Judicial Office, HM Courts and Tribunals Service and the Ministry of Justice. It says it complements the judiciary's existing AI guidance, most recently updated in October 2025. It is not a general rule for every lawyer, expert witness or member of the public who uses generative AI to prepare material for court.
That scope should prevent a common category error. Rules for an approved internal summarisation tool are different from the risks posed by an advocate submitting invented authorities or a litigant relying on generated advice. The judiciary's own tool controls can protect internal work, but they cannot by themselves verify material created outside the court system. Separate professional duties, procedural rules and evidence checks continue to matter.[1]
What would make the policy assessable
The statement promises continuing review as technology, legal requirements, pilot evidence, incidents and risks change. The next useful evidence would be a public evaluation framework: named task categories, representative test sets, human comparators, error severity, redaction and confidentiality failures, time saved, checking burden and performance across languages, disabilities and case types. Results should be tied to the exact model and version because an approval cannot safely be assumed to transfer to a later release.
Confidence would also rise if the judiciary published short impact and incident summaries that protect case confidentiality while showing what was tested, what failed and why a deployment decision changed. Independent scrutiny and participation by court users would help reveal harms that an internal pilot may miss. Until those details appear, the position is consequential because it formalises who decides, who approves and which basic controls apply. It remains a policy foundation rather than proof of safe performance.[1]
What this means for people
- Judges remain personally responsible for reviewing and approving AI-assisted work rather than deferring to a system.
- Lawyers and court users gain a formal assurance that AI is not authorised to decide cases, but the statement does not yet create an inspection or disclosure right.
- Court staff may use approved tools for administrative and document tasks, subject to central approval and enterprise data controls.
Global context
Courts in many jurisdictions are testing summarisation, transcription and document-analysis tools while confronting hallucination, confidentiality and fairness risks. England and Wales' position is notable for combining an explicit non-delegation rule with central approval and three technical conditions. It should not be assumed to govern Scotland, Northern Ireland or courts abroad, whose legal duties, disclosure rules, languages and technical infrastructure differ.
What the evidence does not yet show
- This is the judiciary's own policy statement, not an independent evaluation of its pilots or approved tools.
- The document publishes no product list, pilot denominator, accuracy result, error rate, case-type breakdown or user-impact evidence.
- Human review is required, but the statement does not define verification procedures or minimum performance thresholds.
- The scope covers tools used by or on behalf of the judiciary, not every use of AI by lawyers, witnesses or court users.
- Publication and disclosure rules for evaluations, incidents and use in individual proceedings are not specified.
What to watch next
- Published evaluation methods and results for lower-risk judicial AI pilots
- Disclosure and correction routes when an AI-supported output affects a proceeding
- Version-specific approval, incident reporting and re-testing requirements
- Evidence on accessibility, language performance, bias and the workload of human verification
- Whether any operational use expands beyond the support functions listed in the statement
Living evidence record
Impact record IAI-1J5QNXP
Evidence stage
Announced
Confidence
Developing
Reporting basis
Source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
12 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what Courts and Tribunals Judiciary published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 12 October 2026
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Continue the story
Related reporting
Government & Policy
Governments are using AI—but how many can prove it improves public services?
The OECD reports government AI use in 35 of 36 participating member countries, yet only 10 say they measure financial or non-financial impact. The comparison maps national practice; it does not audit individual systems or prove public benefit.
7 min · 2 sources
Government & Policy
Who sets the rules for genomic AI?
A peer-reviewed review mapped 90 publicly documented national genomics initiatives across 70 countries and territories. Thirty-two reported current or planned AI use, but the researchers found public AI-specific governance in only three programmes.
8 min · 4 sources
Government & Policy
What will the new US AI task force actually decide?
President Donald Trump has publicly announced a federal AI task force led by intelligence chief Jay Clayton. The announcement confirms its four-person leadership and stakeholder remit, but the charter, evidence process, enforcement power and final policy test remain unpublished.
9 min · 6 sources
Reader commentary
Add evidence, experience or a question
No account is required. Reader notes are published after a brief civility, relevance and safety check; disagreement is welcome.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.