Regulation & Ethics
UK Judiciary, Bar Council and Law Society AI Guidance: The 2026 Picture
UK judiciary AI guidance (Oct 2025), Bar Council considerations (Nov 2025) and Law Society material share one rule: whoever signs is responsible for every word.
On this page
Three documents govern AI use in the courts of England and Wales as of September 2026: the judiciary's AI guidance for judicial office holders (updated 31 October 2025), the Bar Council's considerations on generative AI (reviewed 25 November 2025) and the Law Society's guidance, now backed by the SRA's Warning Notice of 17 August 2026. None requires disclosure of AI use to a court. All make the human who signs responsible for every word.
Three documents side by side
| Judiciary | Bar Council | Law Society and SRA | |
|---|---|---|---|
| Document | AI guidance for judicial office holders | "Considerations when using ChatGPT and generative AI" | Law Society guidance and press release; SRA Warning Notice on misuse of AI |
| Date | 31 Oct 2025 (replacing the April 2025 version) | Reviewed 25 Nov 2025 | SRA notice 17 Aug 2026 |
| Status | Guidance issued by the senior judiciary | Bar Council guidance; not formal BSB guidance | Law Society: guidance; SRA: regulatory warning applying the Principles and Code |
| Core rule | Responsible for all material produced in the judge's name | The barrister's core duties apply to AI-assisted work | Understand the tool well enough to review its output critically |
| Confidentiality | No private information into public AI tools; leaks are data incidents | Do not put confidential or privileged material into public tools | UK v SSHD: pasted client letters are in the public domain |
| Verification | Judges check AI output like any other draft | Barristers verify authorities and quotations personally | Every authority and quotation checked against the source |
| Disclosure to court | Not required | Not required | Not required |
The absence of a disclosure rule is deliberate. The English position is that the existing duties to the court and the existing sanctions are sufficient, which is the same conclusion the Illinois Supreme Court reached in its 2025 policy and the opposite of the certification orders that more than 30 US federal districts have adopted; the standing orders page compares them.
What judges are told
The October 2025 guidance is the third iteration and the first to read as if written after real incidents. It permits judicial office holders to use AI for tasks where it helps, such as summarising material or drafting routine text, and then draws three lines. Judges must not enter private or confidential information into public AI tools. Any inadvertent disclosure is to be reported as a data incident. And judges remain responsible for "all material produced in their name", which is the sentence that governs everything else.
That last line matters more than it looks because judges have already been caught out elsewhere. Charlotin's database records 32 decisions worldwide in which the court's own ruling contained hallucinated material. In the United States, Judge Julien Neals (D.N.J.) and Judge Henry Wingate (S.D. Miss.) withdrew orders in July 2025 containing fabricated quotations and non-existent parties; in letters to Senator Grassley in October 2025 they admitted an intern had used ChatGPT and a clerk had used Perplexity, as Reason reported. On 31 August 2026 a Fifth Circuit panel heard Mississippi's request to remove Judge Wingate from a case because hallucinated citations persisted even in his amended order. The English guidance is designed to prevent the equivalent.
For advocates the practical consequence is modest but real. A judge may use an AI tool to summarise your skeleton. The guidance makes the judge responsible for the summary, but it does not stop you writing skeletons whose structure survives summarisation: numbered propositions, one authority per point, quotations set out in full rather than paraphrased. The red-teaming playbook includes a step for running your own skeleton through a summariser to see what a hurried reader, human or machine, would take from it.
What barristers are told
The Bar Council's document is careful about its own status: it is guidance, not BSB rules. That distinction did not help the barrister in Ayinde v Haringey [2025] EWHC 1383 (Admin). The BSB Handbook's core duties, in particular CD1 (duty to the court), CD3 (honesty and integrity), CD6 (confidentiality) and CD7 (competent standard of work), apply to AI-assisted work as to any other, and the Divisional Court referred her to the BSB on that basis. Her denial that she had used AI was found "unpersuasive"; the court held the contempt threshold was met but brought no proceedings because she was junior and already reported.
The considerations document covers the ground you would expect: hallucination, confidentiality and privilege, the limits of a model's training data, and the barrister's personal responsibility for the product. Two points are specific to the Bar. First, the Divisional Court in Ayinde addressed "heads of chambers" directly, so chambers now need a policy and a supervision line in the same way firms do. Second, the pupil and junior end of the profession is where the pressure lands: the barrister in Ayinde was a pupil at the time, and the reasoning that spared her contempt proceedings will not be available to a tenant of ten years.
What solicitors are told
The Law Society's guidance and the SRA's Warning Notice are covered in detail on the SRA notice page. The short version: the SRA received 42 AI-misuse reports between July 2025 and July 2026, cites BCP v A Mother [2026] EWFC 71 (B), Cork v Smith [2026] EWHC 1199 (Ch) and UK v SSHD [2026] UKUT 81 (IAC), and sets the standard that a solicitor must understand a tool well enough to review its output critically. Cork v Smith, in which a Pinsent Masons associate filed an invented "Insolvency Rule 12.37(5)" after the tool had flagged its own possible inaccuracy, is the case every supervising partner should read. The firm was referred to the SRA and bore the extra costs.
How the courts treat AI-assisted submissions
The enforcement mechanism in England and Wales is the Hamid jurisdiction, the High Court's inherent power to regulate the conduct of lawyers appearing before it. Ayinde and Al-Haroun v Qatar National Bank, heard together on 6 June 2025 by Dame Victoria Sharp P and Johnson J, used it to set the baseline. The judgment found five fake authorities in Ayinde and 18 fictitious references among 45 in Al-Haroun, ordered £2,000 plus VAT in wasted costs in each, referred the lawyers to their regulators and listed the available sanctions in ascending order.
| Consequence | Where it has been applied |
|---|---|
| Public admonition | Ayinde; Bandla v SRA [2025] EWHC 1167 (Admin), where fictitious authorities were cited against the regulator itself |
| Wasted costs | Ayinde and Al-Haroun, £2,000 plus VAT each |
| Firm bears extra costs | Cork v Smith, June 2026 |
| Regulatory referral | Ayinde (BSB and SRA); Cork v Smith (SRA) |
| Confidentiality finding | UK v SSHD, Upper Tribunal, 2026: public AI tools put client letters "in the public domain" |
| Contempt and police referral | Listed in Ayinde as available; threshold found met but not pursued |
| Striking off | SRA v Kumar, SDT, decided 25 August 2026: a registered foreign lawyer who defended himself with AI-fabricated citations, then filed a "corrected" version with more, was struck off, per Legal Futures |
The volume is rising. The Law Gazette reported on 18 September 2026 that a litigant had filed a 300-page skeleton argument "generated entirely by ChatGPT". Litigants in person account for most entries in Charlotin's database (1,174 of 2,041 worldwide), and English courts are now receiving machine-written bundles from both represented and unrepresented parties. The casebook tracks the UK entries alongside the US ones.
Ten practical rules for barristers and solicitors
- Treat every AI output as a draft from an over-confident pupil: useful, unsigned, unverified.
- Open every authority the tool cites. Read the paragraph it relies on. Check the quotation word for word.
- Never ask the tool to confirm its own citations; ask a database.
- Put no client, witness or privileged material into a public tool. Enterprise tiers with contractual no-training terms are the minimum; the confidentiality checklist says which tools qualify.
- Keep a note in the file of what was checked and by whom; if a court asks, the answer should take a minute.
- Tell the client, in the client care letter, that AI tools may be used and what safeguards apply.
- If you find an error after filing, write to the court and the other side the same day. Every escalated sanction in the UK and US cases followed a denial or a quiet refiling.
- Supervisors ask whether AI was used before they review, and review accordingly.
- Chambers and firms name someone responsible for AI compliance and train everyone once a year with a live demonstration of a hallucination.
- Read the three documents above once, then read Ayinde in full; the explainer is the short route in.
When the other side's skeleton looks machine-made
A US appellate court has already criticised counsel for failing to spot an opponent's fabricated citations, and the reasoning travels. Use this sequence.
- Run every citation through a database before the hearing; flag any that do not resolve, any whose year or court does not match, and any quotation that cannot be found in the reported text.
- Write to the other side identifying the specific problems and asking them to confirm or withdraw. Give a short deadline.
- If they do not respond, file a short note for the court listing the citations in question with your verification results. State what you checked; do not speculate about how the document was produced.
- At the hearing, ask the court to disregard the unsupported propositions and, if the conduct warrants it, invite the court to consider the Hamid jurisdiction.
- Record the time spent; it is recoverable as costs and the court will want to know.
The litigation guide covers the same process from the drafting side, and the disclosure question page gives the one-paragraph answer for England and Wales alongside the US and EU positions.
Where this leaves you
The English system has chosen responsibility over certification. That is more demanding, not less: a certification can be signed in thirty seconds, while responsibility for "all material produced in your name" follows you into a wasted costs hearing and a regulatory referral. The guidance from the judiciary, the Bar Council and the Law Society is short, free and consistent. Read it, adopt the ten rules, and run the opponent's skeleton through step one before every hearing. Module 6 of the AI Lab for Lawyers course works through Ayinde, Cork v Smith and the US sanctions cases in a single session and ends with the verification drill that would have prevented each of them.
Frequently asked questions
Do UK courts require lawyers to disclose that they used AI?
No. There is no certification or disclosure rule in the courts of England and Wales as of September 2026, unlike more than 30 US federal districts. What exists instead is responsibility: the judiciary's guidance, the Bar Council's considerations and the SRA's notice all make the person who signs a document answerable for its contents, and Ayinde v Haringey lists the sanctions.
Can judges in England and Wales use ChatGPT?
The judiciary's guidance, updated 31 October 2025, permits judicial office holders to use AI tools for suitable tasks but tells them not to enter private or confidential information into public tools, to treat any leak as a data incident, and that they remain responsible for all material produced in their name. Judges cannot delegate judgment to a tool.
Is the Bar Council's AI guidance binding on barristers?
Not formally. The 'Considerations when using ChatGPT and generative AI' document, reviewed 25 November 2025, is Bar Council guidance rather than BSB rules. The BSB Handbook's core duties bind regardless, and in Ayinde v Haringey a barrister who relied on fabricated authorities was referred to the BSB and her denial of AI use was found unpersuasive.
What should I do if the other side's skeleton argument looks AI-generated?
Check every authority it cites before the hearing, because a court expects you to notice fabrications and a US appellate court has criticised counsel who did not. Raise it in correspondence first, then in a short note to the court identifying the specific citations that do not exist or do not say what is claimed. Do not accuse; show.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Regulation & Ethics
The SRA's 2026 Warning Notice on AI Misuse: What UK Solicitors Must Know
The SRA AI warning notice of 17 August 2026 followed 42 misuse reports in a year. Solicitors must understand a tool well enough to review its output critically.
8 min read
Regulation & Ethics
Ayinde v Haringey Explained: The UK's Leading AI Hallucination Judgment
Ayinde v Haringey [2025] EWHC 1383 (Admin) explained: the fake cases, the Divisional Court's sanctions ladder, the duty on heads of chambers and how the SRA built on it.
11 min read
Regulation & Ethics
Court Standing Orders on AI: Where You Must Disclose or Certify
Court AI disclosure rules in 2026: over 30 of 94 US federal districts and 300+ judges require certification, two Ohio judges ban it, Illinois requires none.
10 min read
Practice Areas
AI for Litigators: Research, Drafting, Discovery and the Sanctions Trap
AI for litigators pays off in chronologies, discovery and deposition prep; it costs $110,204 when a brief cites invented cases. Where to gain and the one rule.
8 min read
Playbooks
Red-Team Your Argument with AI: Adversarial Prompts for Litigators and Negotiators
Red-team your legal argument with AI in five rounds: prompts that make the model argue the other side, play a sceptical judge and score your case before filing.
12 min read
Questions
Do I Have to Tell the Court I Used AI? Disclosure Rules by Jurisdiction
Whether you must tell the court you used AI depends on the judge: 30+ US federal districts require it, Illinois does not, England has no duty. How to check.
6 min read