Forty-two. That is how many reports of potential AI misuse the Solicitors Regulation Authority received between July 2025 and July 2026, some “from senior members of the judiciary”. On 17 August 2026 the regulator answered with a formal warning notice, “Misuse of AI”, and the SRA AI warning notice is now the document every COLP in England and Wales is expected to have read.
What makes it worth reading is the case it leans on hardest. The solicitor in [2026] UKUT 81 (IAC) never drafted anything with ChatGPT. He ran a Google search, Google’s “AI Mode” answered with a Court of Appeal authority called Horleston v SSHD, complete with a bench, and he cited it. The case does not exist. He had also pasted client letters and Home Office decision letters into ChatGPT to “improve” them, which the tribunal held placed that material “on the internet in the public domain”.
The message, then, is not “stop using ChatGPT” but that every tool produces output you are accountable for, and that the confidentiality analysis starts before you type.
What the SRA AI warning notice of 17 August 2026 actually says
The central sentence of the notice:
“AI has no separate legal personality; solicitors and regulated individuals who use AI in the course of delivering legal services remain accountable for their work and outputs, regardless of how that work has been prepared.” — SRA warning notice, Misuse of AI, 17 August 2026
Citations: every authority must be verified; “reliance on an output of AI would not be a suitable defence”. Confidentiality: “Both paid for and free-to-use AI tools may not provide the contractual, and technical safeguards needed to maintain client confidentiality”; client data goes in only where it “is not used to train AI models except where explicitly authorised and appropriate to do so, and is not retained longer than necessary”. Supervision: open investigations concern “inaccurate legal citations, supervision and confidentiality”. Accountability: “You remain accountable for all work and advice delivered to clients, whether or not AI has been used in its preparation.”
Aileen Armstrong, the SRA’s Executive Director for Strategy and Policy, summarised it: “the use of AI does not change the professional standards expected of solicitors and law firms.” The same regulator approved Garfield.Law in May 2025 as the first AI-only law firm: the notice is about misuse, not use.
What the SRA expects, mapped to the Principles and the Code
The notice names the existing rules that bite.
| Expectation | Source | What compliant looks like |
|---|---|---|
| Verify every authority and quotation before filing | Principles 1, 2, 4, 5; Code 1.4, 2.1–2.7 | A record of who opened each authority at source; never a “does this case exist?” question put to the same tool |
| Competence and up-to-date skills | Code 3.2, 3.6 | Dated, hands-on training on the tools staff actually use |
| Effective supervision | Code 3.5; Code for Firms 4.2–4.4 | Supervisors know which matters used AI and checked the output as they would a trainee’s |
| Confidentiality and privilege | Principle 7; Code 6.3; Code for Firms 2.3 | Client data only in tools with contractual no-training and retention terms; consumer tiers barred for anything identifiable |
If your AI policy does not cite those paragraphs, it predates the notice; the law firm AI policy template is built around them. The full list in the notice: Principles 1, 2, 4, 5 and 7; Code for Solicitors 1.4, 2.1–2.7, 3.2, 3.5–3.6, 6.3 and 7.2; Code for Firms 1.4, 2.3, 4.2–4.4, 6.3 and 8.1.
Law Society, Bar Council and the judiciary
The Law Society’s “Generative AI – the essentials”, updated June 2026, has the clearest confidentiality rule in any UK document: “If you are using a free, online generative AI service where you have no operational relationship with the vendor other than use, do not put any confidential data into the tool.” It also records Rodney v Gee’z Micro Bar & Pitstop (April 2026), where two solicitors were referred to the SRA over AI-derived citations.
The Bar Council’s guidance, issued 30 January 2024 and updated 26 November 2025 to cover Gemini, Perplexity, Harvey and Copilot, says the same for the Bar. The judicial AI guidance, current as of 31 October 2025, tells judges, clerks and staff “not to enter private information into public AI tools” and, in Lord Justice Birss’s words, that they bear “personal responsibility” for “all material produced in their name”.
Ayinde and Al-Haroun: the Divisional Court’s warning
R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), decided on 6 June 2025 by Dame Victoria Sharp P and Johnson J, is the judgement every later UK document cites.
“Freely available generative artificial intelligence tools, trained on a large language model such as ChatGPT are not capable of conducting reliable legal research. Such tools can produce apparently coherent and plausible responses to prompts, but those coherent and plausible responses may turn out to be entirely incorrect.” — Dame Victoria Sharp P, Ayinde, para 6
Three passages belong in the compliance file. Paragraph 7: those who use AI for research “have a professional duty therefore to check the accuracy of such research by reference to authoritative sources”. Paragraph 9, for managing partners and heads of chambers: “the profession can expect the court to inquire whether those leadership responsibilities have been fulfilled.” Paragraph 82: “promulgating such guidance on its own is insufficient to address the misuse of artificial intelligence.”
UKUT 81: the statement of truth and the “public domain” finding
UK and R (Munir) v SSHD [2026] UKUT 81 (IAC), promulgated on 17 November 2025, added two things. First, the Upper Tribunal’s judicial review claim form now requires a statement of truth that every authority cited “(a) exists; (b) may be located using the citation provided; and © supports the proposition of law for which it is cited”. Second, at paragraph 21:
“to put client letters and decision letters from the Home Office into an open source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege.” — Upper Tribunal (Immigration and Asylum Chamber), [2026] UKUT 81 (IAC), para 21
The tribunal contrasted ChatGPT with “closed source AI tools … such as Microsoft Copilot”, meaning the work-tenant version, not the consumer app (see Microsoft Copilot for lawyers). It also held that supervisors “must ensure that fee-earners under their supervision are aware of the dangers of using non-specialist Artificial Intelligence (AI) for legal research and drafting”.
UKJT: liable for failing to use AI
In July 2026 the UK Jurisdiction Taskforce, chaired by the Master of the Rolls, Sir Geoffrey Vos, published its Legal Statement on Liability for AI Harms. Paragraph 67:
“It is important to be aware of the possibility that a professional could also be liable for failing to use AI for a task when a professional exercising reasonable care and skill would have done so.” — UK Jurisdiction Taskforce, Legal Statement on Liability for AI Harms, para 67 (July 2026)
That is ordinary negligence law. ABA Formal Opinion 512 says the same in American: “it is conceivable that lawyers will eventually have to use them to competently complete certain tasks.”
Hill Dickinson’s 32,000 hits, Mishcon’s reversal and 61% adoption
In February 2025 Hill Dickinson found more than 32,000 ChatGPT hits in seven days, more than 3,000 on DeepSeek and around 50,000 on Grammarly, and blocked general access. The Information Commissioner’s Office replied via the BBC: “the answer cannot be for organisations to outlaw the use of AI and drive staff to use it under the radar.” Mishcon de Reya restricted ChatGPT in March 2023 while advertising for a “GPT Legal Prompt Engineer”; on 31 July 2025 it rolled Legora out to all fee earners, with a super-user noting that “it remains essential to verify Legora’s output in much the same way as if a trainee had produced the work.” Governed access beats prohibition, because prohibition produces the 32,000 hits you cannot supervise (shadow AI in law firms).
By September 2025, 61% of UK lawyers used AI in day-to-day work, up from 46% in January 2025 (LexisNexis). Damien Charlotin’s hallucination database listed 69 UK decisions among 2,039 on 12 September 2026, collected in AI hallucination cases in the UK, Germany and Europe. Sixty-nine is small against 61% adoption. The SRA needed only 42 reports to act.
A compliance checklist for COLPs
- Tool register, including Copilot, DeepL and search engines with AI answers; the UKUT 81 solicitor’s tool was Google.
- Confidentiality tiering. Consumer tiers: no client-identifiable data. Commercial tiers: only with the contractual terms the SRA lists.
- The three tests on every authority, checked at source, never by asking another AI (how to verify AI legal citations).
- Supervision record: which matters used AI, on which tool, reviewed by whom (Code 3.5).
- Hands-on, dated training on the actual tools (Code 3.6; Ayinde para 82).
- Incident procedure: correct the record and tell the court first (found a fake citation: what to do).
- The negligence mirror: whether a competent peer would now use a tool for the task (UKJT para 67).
Three prompts do the daily work; the anonymisation routine is in the prompt library.
Below is a draft [skeleton argument / letter of advice] under English law. Extract every authority it cites into a table: Authority as cited | Where it appears | Proposition cited for | (a) Exists: NOT YET VERIFIED | (b) Locatable by this citation: NOT YET VERIFIED | (c) Supports the proposition: NOT YET VERIFIED.
Do not fill in the verification columns and do not add, correct or "improve" any citation. I will open each source and complete the columns by hand.
Draft:
[paste]I am a solicitor in England and Wales. Do not ask for, and I will not give, any names, dates, reference numbers or facts that identify a matter.
Explain the general framework that applies when [a local authority decides that an applicant is not in priority need under Part 7 of the Housing Act 1996]: the provisions in play, the questions a decision-maker must answer in order, and the points on which challenge usually turns.
Cite no cases. Tag every statutory provision [VERIFY]. If a point is contested or has changed recently, say so rather than guessing.Summarise this conversation as a supervision note for a solicitor's matter file, plain text, with these fields: Date | Tool and tier used (e.g. Microsoft Copilot, work tenant) | Task | Materials supplied (types only, no client data) | Outputs relied on | Authorities or facts in the output, each marked "verified at source by [initials] on [date]" or "NOT YET VERIFIED" | Outputs discarded and why.
Record process, not the legal substance of the advice.Where to go next: AI ethics rules by jurisdiction puts the SRA beside the ABA, the CCBE and the German and Austrian bars; the other ethics and regulation guides cover disclosure, court orders and insurance. To learn the tiering and verification routine on your own documents, the four live sessions of AI Lab for Lawyers are the place; participants from Simmons & Simmons, Taylor Wessing and Withers have already sat through them.
Frequently asked questions
What does the SRA say about generative AI?
The SRA's warning notice of 17 August 2026 says AI 'has no separate legal personality' and that solicitors 'remain accountable for their work and outputs, regardless of how that work has been prepared'. It expects every cited authority to be verified, states that 'reliance on an output of AI would not be a suitable defence', and limits client data to tools where it is not used for training, not accessed by unauthorised third parties and not retained longer than necessary.
Can UK solicitors use ChatGPT?
Yes, within limits. Neither the SRA notice nor the Law Society bans it. The Law Society says that with a free online service where you have no relationship with the vendor beyond use, 'do not put any confidential data into the tool', and the Upper Tribunal held in [2026] UKUT 81 (IAC) that pasting client letters into ChatGPT places them 'in the public domain'. Use consumer tiers for abstract questions only, and commercial tiers with no-training terms for client work.
What did the court decide in Ayinde?
In Ayinde v Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), the Divisional Court held that freely available tools such as ChatGPT 'are not capable of conducting reliable legal research', that lawyers have a professional duty to check AI research against authoritative sources, and that managing partners and heads of chambers will be asked whether they fulfilled their leadership duties. It listed sanctions from admonition to contempt and police referral, and referred the lawyers to their regulators.
Do UK courts require AI disclosure?
Neither the SRA notice, the Law Society guidance nor Ayinde imposes a general duty to declare AI use in filings, unlike the Federal Court of Canada's first-paragraph declaration. The Upper Tribunal's judicial review claim form does require a statement of truth that every authority exists, can be located by its citation and supports the proposition cited, and the Civil Justice Council consulted on AI in court documents in February 2026. Accuracy, not disclosure, is the enforced duty.
Could a solicitor be liable for not using AI?
Possibly. The UK Jurisdiction Taskforce's July 2026 Legal Statement on Liability for AI Harms says at paragraph 67 that 'a professional could also be liable for failing to use AI for a task when a professional exercising reasonable care and skill would have done so'. That is ordinary negligence law applied to a new tool: the standard is what a competent peer would do, which increasingly includes using validated tools for review and checking.