In partnership with
The Civil Justice Council's final report on AI and court documents is out, and I think most litigators will read the headline (no AI declaration for skeleton arguments or statements of case) with some relief before reaching the part which affects them most: witnesses of fact would have to declare that AI did not generate or influence the substance of their evidence. A family court judgment on an expert report generated with Copilot shows the same concern already at work. Elsewhere this week, the CPS relied on a case which does not exist, and a new report warns about what agreeable chatbots may do to family disputes.
AI in Practice
The CJC wants witnesses to declare that AI did not shape their evidence
On 29 October the Civil Justice Council published its final report on the use of AI for preparing court documents, with 15 recommendations from a working group chaired by Sir Colin Birss, who the Gazette reports has been sworn in as Master of the Rolls this week (Law Society Gazette). Lawyers' own documents attract little new regulation. There would be no AI declaration for statements of case, skeleton arguments or other written advocacy, no requirement to list the tools used and no need for the court's permission to use them, although the CPR would be amended so that "the specific legal professional who takes responsibility for the document is identified" (Burges Salmon). Disclosure waits on the separate review of PD 57AD, and the Civil Procedure Rule Committee would draft a practice direction setting out the existing expectations on accuracy, verification, responsibility and sanctions.
Evidence is treated differently. As reported by Legal Futures, a witness of fact would have to declare that the statement was not prepared using an AI tool which "generated or influenced its substantive content", with the focus on "the result of such use, not on the use itself or the tool that produces the result" (Legal Futures). That would apply to litigants in person as well as represented parties, and to evidence "whether prepared for trial or otherwise", with statements prepared and signed by a legal representative (other than as a witness of fact) left to existing professional obligations. Experts would make a similar declaration where AI generated or influenced the substance of their report, with enough detail to allow their evidence to be "scrutinised and tested effectively by opposing parties and the court". According to Hogan Lovells Cadwalader, the report compares AI which alters factual evidence to "witness coaching" (Hogan Lovells Cadwalader).
This author's view is that the declaration would be the witness's, but the risk would probably sit with whoever prepared the statement. Many statements begin as an attendance note or an interview transcript, and if a fee earner asks a tool to turn that into a first draft, the witness may be asked to sign a declaration they cannot truthfully give, about a step they never saw. The line is not yet clear and Hogan Lovells Cadwalader points out that "a transcript of a witness interview may then be tidied, condensed or reorganised using AI", and asks when that becomes influence on the evidence, and it suggests that even spellcheck is not always straightforward. Anyone who has worked under PD 57AC will recognise the concern about a witness's own words, but this would reach well beyond the Business and Property Courts.
None of this is in force, and the rule committee has to act before anything changes; one commentary notes that the report sets no timetable (DMD Costs). It would still be sensible to find out now how statements are actually being prepared in your team, because any file you open today may be running when the declaration arrives, and it may be difficult to reconstruct how an early draft was produced.
Takeaways
Act: Ask each litigator and paralegal whether AI has touched any stage of a witness statement, from transcript to final draft. Until the wording is settled, set a working rule that AI does not generate or rewrite the substance of a witness's evidence, and record any use on the file.
Watch: The Civil Procedure Rule Committee's practice direction and CPR amendments, the promised guidance for litigants in person (which the report wants considered "as a matter of urgency"), and any joint guidance from the legal regulators.
Risk: A witness who signs a declaration without knowing that an AI tool shaped the draft, and a cross-examination which exposes it.
Read: CJC final report, the CJC project page, and the reporting at Legal Futures
On your radar
A family judge would not give significant weight to an expert who generated a report with Copilot: Not this week's news, but newly reported and not covered here before: in Cornwall Council v M [2026] EWFC 306, handed down at the end of September, HHJ Flexman considered kinship assessments by an independent social worker, who accepted in oral evidence that the whole of one assessment was AI generated from her notes. One report still contained wording saying that a passage "is more likely to withstand cross-examination", and the judge found that the inclusion of Copilot's own summary showed "there has not been adequate human oversight of the contents of the report". The judge accepted that there is "a sliding scale of use from what is acceptable to what is not", and was concerned that "the use of AI to generate an entire report from notes strays into the latter". She did not "feel able to place any significant weight" on the evidence, though for wider reasons as well, including a lack of rigour. Why it matters for UK lawyers: this is a Family Court decision about weight rather than a new rule, but it shows what the CJC's proposed expert declaration would be testing. The expert's written disclosure said that Copilot "did not generate evidence, undertake analysis, or form conclusions", and it was only in oral evidence that the extent of its use emerged, although she maintained that the contents were her professional opinion. Add a question to your letters of instruction asking what AI tools the expert uses, for what, and how the output is checked, and read the reports you receive for leftover tool text. (National Archives, gunnercooke)
The CPS relied three times on a case which does not exist: In LYC v Director of Public Prosecutions [2026] EWHC 2778 (Admin), a judicial review of a decision to continue a youth prosecution, the written record of decision referred three times to "R v J [2024] EWCA Crim 3529". The parties agreed that the case does not exist, and the prosecutor said in a witness statement that she had meant R v JTB [2009] UKHL 20. The Divisional Court (Holgate LJ and Wall J) called the explanation "unsatisfactory", but on whether it mattered to the outcome said "Our firm answer is no", because the real question was whether the principles applied were wrong in law. The claim failed. The judgment does not mention AI or say how the citation arose, and nothing here should be taken as suggesting that a tool was involved. Why it matters for UK lawyers: a phantom authority can appear in a decision you are challenging as well as in a document filed against you, and this case shows that catching one may not win the point on its own. Check every authority cited in a decision letter or an opponent's document against the primary source, and if one does not exist, frame the challenge around the legal proposition it was cited for. (National Archives, Law Society Gazette)
"Sycophantic" AI may be inflaming family disputes, a new report warns: Dr Natalie Byrom's report for the Nuffield Family Justice Observatory, Delegated by Default?, dated 2 November, argues that government is leaving digital and AI-mediated justice to the private market, and that general-purpose tools such as ChatGPT can fall outside legal services regulation, leaving people with limited routes to redress. As reported by the Gazette, it warns that sycophantic AI (the tendency covered in Wednesday's Jargon Buster) "may affirm and enflame vexatious litigants and domestic abusers" while helping them to produce convincing documents at scale. It makes 26 recommendations, including a National Commission on the Regulation of AI in Justice. On the reporting so far, the concern about abuse is a warning about risk and not a measured effect, and the recommendations are not government policy. Why it matters for UK lawyers: family practitioners may increasingly face a well-presented application from a party whose chatbot has told them they are right. If you act for someone on the receiving end of a high volume of polished documents, keep a schedule of what has been served and when, and consider raising the pattern with the court at the directions stage. (Nuffield Family Justice Observatory, Law Society Gazette)
Ad Break
In order to help cover the running costs of this newsletter, please check out the advert below. In line with my promises from the start, adverts will always be declared.
9 out of 10 clients get approved for Cherry Plans
At Cherry, we provide payment plans to 90% of pet parents. Plus, we make applying easy – with a 35-second online application they can submit right in the treatment room, and a soft credit check that doesn’t ding their credit history.
Don’t let cost get in the way of care - with Cherry.
* These figures come from Cherry's provider case studies, so individual results vary. Details at withcherry.com/provider-reviews. Payment options through Cherry are issued by the following financing partners: withcherry.com/financing-partners
For Review
Rising caseloads in judicial review: a question of AI? (Sam Guy and Joe Tomlinson, UK Constitutional Law Association)
A post from 30 October which tests the suspicion, recorded in the minutes of the Administrative Court User Group, that litigants in person are filing more claims with AI help. There were 2,630 judicial review applications in the first half of 2026, which on a straight-line projection would be 36% above 2025, but the authors find that permission rates and "totally without merit" certifications do not yet clearly show the fall in quality one might expect, and they describe a "mixed picture" in which the link to AI is unproven. It is a useful corrective to confident claims about AI-driven caseloads, and public law practitioners should read it before repeating them in submissions or to clients.
Read or listen: UK Constitutional Law Association
Is legal advice developed by or with an AI system legally privileged? (Phillip Kelly, Rachel Tookey and Crystal Sawyers, DLA Piper)
An update from 2 November to the authors' 2025 article, which argues that there is no standalone "AI privilege": advice generated by an AI system and given directly to a non-lawyer is not privileged, but privilege can attach to advice developed with AI where a lawyer directs and supervises the tool. It covers Munir in the Upper Tribunal and the US decision in Heppner, and warns that prompts and transcripts may be scrutinised, as the authors say happened in Cork v Smith. In-house teams in particular should use it to check whether a paid or enterprise tool's terms really protect what they put into it, since price alone is no safeguard.
Read or listen: DLA Piper
Copilot for Legal: interview with Carina Negreanu of Microsoft (Artificial Lawyer)
A follow-up to the launch covered here on 23 October, adding some detail: Copilot for Legal will use both OpenAI and Anthropic models, and the capability "will automatically make itself known to the user if it detects legal documents". The interview does not mention the UK or English law, and the figures on accepted edits are Microsoft's own. Read it if your firm has not yet decided whether to leave the plugin switched on, and add the question of which model is handling your documents to your list for Microsoft or your IT provider.
Read or listen: Artificial Lawyer
Practice Prompt
Try the below prompt to map where AI could touch the witness statements and expert reports your team prepares, and to draft a working protocol ahead of the declarations proposed by the Civil Justice Council. Ensure you fill in context and constraints and other aspects marked with {}. Remember to adhere to the Golden Rules and do not upload confidential or privileged information to public tools.
You are assisting a head of litigation, supervising solicitor or COLP at a law firm in England and Wales who wants to review how AI could affect the evidence the team prepares, ahead of the Civil Justice Council's final report of 29 October 2026 being implemented. As publicly reported, and not yet in force, the report recommends that witnesses of fact declare that their statement was not prepared using an AI tool which generated or influenced its substantive content; that experts declare AI use which generated or influenced the substance of their report, with enough detail to allow scrutiny; that statements prepared and signed by a legal representative, other than as a witness of fact, remain subject to existing professional obligations; and that the approach should look at the function a tool performs and the risk that creates, not at categories of tool. Your task is to produce a workflow review and a draft working protocol. You are not concluding that any past statement or report is defective, and you are not concluding that the team is compliant.
Context to apply:
- The team and the work: {e.g. two solicitors, a trainee and a paralegal doing county court civil litigation; say whether any work is in the Business and Property Courts or the Family Court}
- How a witness statement is currently produced, step by step: {e.g. interview, recording or attendance note, first draft, witness review, statement of truth}
- The AI tools available at each step, approved or not: {e.g. transcription tool, Microsoft Copilot, firm-licensed drafting tool, personal chatbot accounts; say "unknown" if nobody has asked}
- What those tools have actually been used for on statements so far: {describe without client names, or say "unknown"}
- How the firm instructs experts and what the letter of instruction says about AI: {quote or summarise, or say "nothing"}
- How clients and witnesses are told about AI use in their own preparation: {describe, or say "they are not"}
- The firm's AI policy and supervision arrangements: {summarise, or say "none in writing"}
Produce the review under these headings:
1. Workflow map
Set out each step in preparing a witness statement, and for each step state whether AI is used, might be used, or is excluded. Mark separately what has been confirmed and what has been assumed.
2. Function and risk
For each use of AI identified, describe what the tool actually does (for example transcribing, correcting spelling, condensing, reordering, drafting, suggesting content) and rate the risk that it generates or influences substantive content as low, unclear or high, with a one-line reason. Treat anything you cannot place with confidence as unclear, and do not treat a tool as safe because of its name or its vendor.
3. The declaration test
For each step rated unclear or high, ask whether the witness could truthfully sign the reported declaration if that step were used on their statement, and what they would need to be told first. Identify any step which the witness would not otherwise see.
4. Witnesses using AI themselves
Set out how the team will find out whether a client or witness has used a chatbot on their own account, draft a short plain-English question and explanation the team can use, and state what should happen if the answer is yes.
5. Experts
Draft a short paragraph for letters of instruction asking what AI tools the expert uses, for which tasks, how the output is checked, and whether any AI has generated or influenced the substance of the report. List what to look for on receipt, including leftover tool text and language which does not match the expert's oral or earlier written work.
6. Draft working protocol
Draft a protocol of no more than ten short rules for the team, covering permitted and prohibited uses on witness evidence, how use is recorded on the file, who signs off, and what happens if use is discovered late.
7. Action list
Give no more than six actions, each with an owner and a date, in order of priority.
8. What this review cannot tell you
Flag every point resting on an assumption about the team, the tools, the Civil Justice Council's report or the law, say what the firm would need to establish to replace it with a fact, and list the points on which the final wording of any rule or practice direction may change the answer.
Constraints:
- {Add firm-specific constraints, for example a client who has prohibited AI use, an insurer requirement, a trial date, or a limited training budget.}
- Apply the law of England and Wales, the Civil Procedure Rules as they currently stand, and the SRA Standards and Regulations. Treat the Civil Justice Council's recommendations as recommendations only, and where you are unsure what the report says, say so and tell the user to check the report itself.
- Do not invent rules, practice directions, declaration wording, case law or product features. If you cannot verify something, mark it as unverified.
- Do not suggest wording which helps a witness or expert avoid making a declaration they ought to make, and do not suggest changing the substance of any witness's evidence.
- Do not conclude that any statement, report or process is compliant. The purpose is to surface risks and open questions.
- Write in plain English and keep each finding short enough to act on.
- This is a planning aid, not legal or regulatory advice. The firm and its supervising solicitors remain responsible for the evidence they prepare and for any use of AI in client work.How did we do?
Hit reply and tell me whether AI has found its way into how your team prepares witness statements, and where you would draw the line.
Serhan
UK Legal AI Brief
Disclaimer
Guidance and news only. Not legal advice. Always use AI tools safely.
Recommended Newsletters
Below are a few newsletters that I recommend, for various reasons. Check them out!
