England and Wales: the judiciary’s AI guidance

Who it binds

The guidance applies to all judicial office holders under the responsibility of the Lady Chief Justice and the Senior President of Tribunals, and to “their clerks, judicial assistants, legal advisers/officers and other support staff”. It is published online, in the judiciary’s words, “to promote transparency, open justice and public confidence”.

It is the third version. December 2023 was the first; April 2025 replaced it; 31 October 2025 replaced that. Sir Geoffrey Vos, Master of the Rolls and Head of Civil Justice, has described AI as an important, innovative and useful tool that must be treated with appropriate care.

What it tells judges about the tools

“Public AI chatbots do not provide answers from authoritative databases. They generate new text using an algorithm based on the prompts they receive and the data they have been trained upon.”

And the point that matters most to anyone practising outside the United States:

Their “view” of the law is “often based heavily on US and historic law, although some do purport to be able to distinguish between that and the law of England and Wales.”

Qatar’s practice direction makes the same observation at paragraph 10. Two judiciaries, independently, have identified the same failure: a model trained mostly on American material answering a question of local law with American confidence.

Confidentiality, stated as bluntly as it gets

“Do not enter any information into a public AI chatbot that is not already in the public domain. Do not enter information which is private or confidential. Any information that you input into a public AI chatbot should be seen as being published to all the world.”

The guidance adds that public chatbots remember every question asked and every input given, that the information is then available to answer other users’ queries, and that chat history should be disabled where the option exists. The October refresh also, for the first time, tells judicial office holders where to report an inadvertent disclosure as a data incident. That is a judiciary planning for the leak rather than forbidding it.

Deepfakes, and hidden white text

On forged material the guidance is calm: courts “have always had to handle forgeries, and allegations of forgery, involving varying levels of sophistication”, and judges should be aware of the new possibility deepfakes present. Then it names something no other judicial instrument on this record has:

“Another form of fake material of which you must be aware is so called ‘white text’, which consists of hidden prompts or concealed text inserted into a document so as to be visible to the computer or system but not to the human reader.”

The glossary defines it precisely: “Text formatted to be invisible to human readers (e.g. white font on a white background) but still detectable by computers. It can be used to manipulate search engines or large language models by embedding hidden instructions or keywords.”

This is prompt injection, arriving in a filed document, named by a judiciary. The guidance draws the right conclusion from it: the possibility “underscores the importance of judicial office holders’ personal responsibility for anything produced in their name.” The same logic applies to any lawyer who feeds an opponent’s document into a summarising tool.

The position statement of 9 October 2026

On 9 October 2026 the judiciary published a Position Statement on the Judicial use of Artificial Intelligence, issued by Sir Colin Birss, President of the Business and Property Division and Lead Judge for Artificial Intelligence. It sits beside the guidance above rather than replacing it, and it sets out how AI is used by and for judges, what safeguards apply, and who approves new uses.

“Every AI output used in judicial work must be reviewed, verified and approved by the responsible judicial office holder. The judicial office holder remains fully accountable.”

Only approved tools may be used for judicial purposes, and they are not listed, because of the pace of change. Each must have three features: UK data residency, enterprise data protection, and disabled web access. AI tools “must not determine the outcome of a case or displace judicial reasoning, analysis or discretion.” Every new use, product or pilot goes to one centralised steering group chaired by the Lead AI Judge, which the statement calls “the sole route” for oversight of developing, piloting and deploying AI tools.

The pilots so far have tested summarisation, transcription, redaction, writing in accessible language, administrative and leadership tasks, chronology-building and document analysis, including how much human checking the outputs need. Training runs through the Judicial College.

The standard a judge is now held to is the one practitioners have been sanctioned for missing: every output reviewed, verified and approved by a named person, who stays accountable for it.

Litigants, witnesses and AI: three decisions of 2026

Hancox v Sutherland, [2026] EAT 139, 17 September 2026. A litigant in person filed a 300-page skeleton argument created using ChatGPT. The Employment Appeal Tribunal used the case to set out guidance on AI for everyone who comes before it:

“Whilst the use of AI is not generally prohibited for tasks such as preparing skeleton arguments, all litigants must take personal responsibility for any document submitted to the Tribunal and it must be carefully checked in the ways I have described, as thoroughly as the author is reasonably able.”

R v FGD, [2026] EWCA Crim 918, 4 June 2026. A complainant had used AI to prepare to give evidence, including mock cross-examination questions and suggested answers: “I had put stuff into AI to help me prepare myself for the court”. The trial judge stayed the prosecution as an abuse. The Court of Appeal reversed the ruling and directed a retrial, noting that, unlike witness training behind closed doors, “all that the AI ‘coach’ provided to C was fully recorded and could be placed in its entirety before the jury”. The Employment Appeal Tribunal in Hancox cites it for the rule that witness coaching by AI, as by other means, is not permitted.

Godwin v Godwin, [2026] EWHC 923 (Ch), 22 April 2026. Litigants in person used ChatGPT on their witness statements. Without the first drafts, the judge said: “I cannot be sure that the Defendant’s, or Ms Godwin’s, witness statements are in their own words”, and approached the evidence cautiously.

In each case the question was what the AI had contributed and what the person had checked. Where the AI’s part was recorded, as in FGD, the court could deal with it; where it was not, as in Godwin, the evidence carried less weight.

The practitioner side: Ayinde

Guidance to judges is not a rule binding filers, so the enforcement route in England and Wales has been the court’s inherent supervisory jurisdiction. Ayinde v Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin) is the leading authority: twenty-three nonexistent citations across two matters, a £2,000 wasted costs order that stood against counsel, a Bar Standards Board referral and two Solicitors Regulation Authority referrals. The court named the sources that count: official government databases, The National Archives and the published Law Reports.

Supervision: the Upper Tribunal, September 2026

UI-2025-002601, Upper Tribunal (Immigration and Asylum Chamber), Lavender J (President) and Upper Tribunal Judge Keith, decision issued 21 September 2026. The grounds of appeal and the skeleton argument were produced by a paralegal using ChatGPT. They cited fake cases, quoted passages that are not in the judgments cited, and gave a quotation from the First-tier Tribunal’s decision that the Tribunal called “pure invention”. The solicitor’s supervision of the paralegal “did not begin to mitigate the risk of filing seriously misleading submissions”. Two individuals reported themselves to the Solicitors Regulation Authority. The appeal failed.

The regulator: the SRA warning notice of 17 August 2026

On 17 August 2026 the Solicitors Regulation Authority issued a warning notice on the misuse of AI, relevant to all firms and individuals it regulates. It is a warning notice, not a new rule: it states how existing obligations apply when AI is in the work. Solicitors remain accountable for all work and advice delivered to clients whether or not AI was used, citations must be genuine, relevant and verifiable, and confidential client material entered into AI tools without safeguards risks permanent waiver of privilege. The notice says in terms that failing to have proper regard to it risks disciplinary action.

Read in full: the SRA warning notice on the misuse of AI, verified against the SRA’s published text.

The professional body: the Law Society practice note

The Law Society of England and Wales practice note, Using artificial intelligence in legal practice, puts the same duty in its own words: “Named regulated solicitors retain accountability for outputs.” Outputs from generative AI need appropriate human verification, oversight and professional judgement, and the note names confidential client information entered into AI systems without contractual and technical safeguards as a risk.

Scotland, Northern Ireland and Ireland

• ScotlandThe Scottish Courts and Tribunals Service uses natural language processing and generative AI for transcription, summarising and translation, and to support online civil forms. Lord Pentland, the Lord President, has said AI is entering the courtroom to stay. We have found no published practice note binding practitioners, and say so rather than assume one exists.
• Northern IrelandNo separate published instrument located.
• IrelandA different jurisdiction entirely, and it has moved. The Courts Service issued guidance on AI in court proceedings on 12 February 2026, and the Law Society of Ireland published Practice Guidance on Generative AI on 20 May 2025. Notably, the Courts Service reported in June 2026 that its judicial Case AI Assistant proof of concept was not considered fit for purpose.

Sources

• Artificial Intelligence (AI) Guidance for Judicial Office Holders, 31 October 2025 (PDF), read in full
• Courts and Tribunals Judiciary, guidance page
• Position Statement on the Judicial use of Artificial Intelligence (AI), Courts and Tribunals Judiciary, 9 October 2026, read in full
• Hancox v Sutherland & Ors [2026] EAT 139, read at Find Case Law
• R v FGD [2026] EWCA Crim 918, read at Find Case Law
• Godwin v Godwin [2026] EWHC 923 (Ch), read at Find Case Law
• Ayinde v Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin)
• Upper Tribunal (Immigration and Asylum Chamber), UI-2025-002601, decision issued 21 September 2026, read in full
• Law Society of England and Wales, practice note, Using artificial intelligence in legal practice, read in full at lawsociety.org.uk
• Scotland and Ireland entries are from reporting and official statements, not from an instrument we have read in full, and are marked as such above.

Related

• AI rules for lawyers, country by country
• Europe: the EU AI Act, the CCBE guides and Germany
• Qatar, whose president is a former Lord Chief Justice of England and Wales
• Australia

Checked against the judiciary’s own PDF and, for the position statement, the statement as published. Information, not legal advice, and not advice on the law of England and Wales.