The verification debate has now moved from the same old story of advocate's citations to a witness's evidence. This week's lead is a Court of Appeal decision on what happens when a witness prepares for cross-examination by asking an AI tool to predict the questions and suggest the answers, and whether that crosses the line into coaching. It arrives just as the Civil Justice Council is weighing a similar question for witness statements, so the theme running through the issue is the integrity of a witness's own account once AI is in the room.
When a witness prepares with AI: the Court of Appeal on coaching
The Court of Appeal (Criminal Division) has handed down a decision that every litigator should read, whether or not they touch criminal work. In R v FGD [2026] EWCA Crim 918, the sole prosecution witness had used an AI tool to prepare for giving evidence, feeding in an account and asking the tool to produce a summary of their recollection, a list of the questions they were likely to face in cross-examination, and suggested answers. The trial judge treated this as prohibited witness coaching and stayed the proceedings as an abuse of process. The prosecution appealed the terminating ruling under section 58 of the Criminal Justice Act 2003, and the Court of Appeal allowed the appeal, so the trial can proceed.
The reasoning is the interesting part. The Court did not treat AI preparation as harmless, and it repeated the settled principle that coaching a witness on their evidence, by any means, is not permitted. What saved the trial was a point about evidence rather than about AI. The danger of conventional coaching is that it happens behind closed doors and cannot be reconstructed, whereas the witness's exchange with the tool was fully recorded and could be put before the jury in its entirety, or reduced to agreed facts. In other words, the very feature that makes AI use worrying (that a witness can rehearse their account against a machine) also makes it visible in a way a private conference with a friend or family member is not.
The Court went further than deciding the case in front of it. It suggested that the National Police Chiefs' Council, the Crown Prosecution Service and the Criminal Procedure Rule Committee consider how the use of AI in criminal proceedings should be addressed and regulated. That is an unusually direct invitation from the Bench for the rule-makers to catch up, and it sits alongside the civil side of the same problem. The Civil Justice Council is exploring proportionate transparency for witness statements, which may include a declaration that AI was not used in their preparation, with its final report expected later in 2026.
This author's reading is that the practical exposure here is wider than the criminal Bar. Any litigator who takes a witness statement now has to reckon with the possibility that the witness has already run their account past a chatbot before the first meeting, and that whatever the tool produced may be disclosable and may go to the weight a court gives the evidence. The safe assumption is that a witness's AI use will come out, so the sensible response is to ask about it, record it, and warn against it at the start, rather than discover it under cross-examination. Reporting restrictions apply to the case, so the usual caution about identifying anyone involved holds.
Takeaways
Act: Add a plain question about AI use to your witness-proofing process, and warn witnesses in writing not to use AI tools to prepare their account or rehearse answers, before they give their first statement.
Watch: Whether the CPS, the National Police Chiefs' Council and the Criminal Procedure Rule Committee take up the Court's invitation, and whether the Civil Justice Council's final report lands on a declaration for witness statements.
Risk: A witness's AI exchange may be disclosable and may undermine the authenticity of their evidence, so an undiscovered chatbot session is a problem you want to find before your opponent does.
Read: The National Archives, R v FGD [2026] EWCA Crim 918, and the commentary at Law Society Gazette
On your radar
Anthropic's Opus 5 lands, and it is already in Harvey: Anthropic released Claude Opus 5 on 24 July, pitched as cheaper than its predecessor with a toggle between cost and capability, and Harvey has confirmed the model is now live in its platform, reporting gains on its own legal benchmark and around a quarter fewer tokens for comparable work. Why it matters for UK lawyers: the engine underneath the tools many firms have just bought has changed again, which affects both quality and the cost per matter that increasingly gets passed through to clients. If your firm uses a platform built on a third-party model, ask your vendor which model version you are on and what changes when they switch. (Anthropic, Harvey, TechCrunch)
The SRA weighs in on agentic AI: In a blog tied to its July Innovate event in Bristol, the Solicitors Regulation Authority set out how it is thinking about agentic AI, the systems that chain tasks together through a coordinating agent, and the consistent advice from speakers was to start with client need rather than the technology. Why it matters for UK lawyers: this is the regulator signalling interest in a class of tools that acts, rather than just drafts, which raises sharper supervision and accountability questions than a chatbot does. If you are piloting anything that takes steps on its own, write down who is accountable for each step before it goes near a client matter. (Solicitors Regulation Authority)
The civil courts edge towards an AI declaration for witness statements: Alongside this week's criminal decision, the Civil Justice Council is reported to be considering whether litigators should have to declare that AI was not used in preparing a witness statement, as part of a wider move towards proportionate transparency. Why it matters for UK lawyers: this is the one genuinely open question left by the Council's interim position covered a fortnight ago, and it would put a positive obligation on the person taking the statement. Start recording now whether and how AI touched any witness statement, so a future declaration is a matter of checking your file rather than reconstructing it. (Legal Futures)
Should clients expect price cuts from legal AI?: A piece in Artificial Lawyer argues that AI will only lower what clients pay if legal work is treated as a utility rather than a luxury good, and that for much of the market the saving will show up as firm margin rather than a lower bill. Why it matters for UK lawyers: clients are starting to ask the question directly, and "we use AI, so why is the bill the same" is a conversation worth preparing for rather than improvising. Decide your firm's line on whether AI efficiency is passed on, absorbed, or reflected in scope before a client raises it. (Artificial Lawyer)
In case you missed it: a structured way to take instructions from overwhelmed clients: This author published a piece this week on a new Client Instruction Schedule skill, now on Lawve, which breaks a struggling client's case into manageable, issue-by-issue questions so you can gather their evidence and instructions without swamping them. Why it matters for UK lawyers: it is a practical companion to this week's theme of capturing a client's own account in a controlled, documented way, and it is built for the litigant who is drowning in their own paperwork. Read it if you deal with clients who cannot get their story onto the page. (UK Legal AI Brief)
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.
How owning AI deployment expands your career
Across product, ops, and CX teams, a new kind of role is taking shape: the person responsible for making AI actually work, day to day. In this roundtable, three people living this shift share what it's really like: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. You'll hear how they carved out these roles, what the job looks like across industries, the skills they'd hire for, and the challenges they're tackling right now.
Watch the full conversation on demand.
For Review
AI as witness coach? Court of Appeal draws the line (Slaughter and May, The Lens)
A clear analysis of this week's lead decision, working through why the recorded nature of the AI exchange mattered and what the ruling means for how witnesses are prepared. If you take witness evidence, this is the version to read closely before you next sit down with a witness.
Read or listen: Slaughter and May, The Lens
Civil and criminal courts put focus on witnesses' use of AI (Legal Futures)
A useful piece that reads the criminal decision and the Civil Justice Council's witness-statement work as two halves of the same problem. Worth reading if you want the civil and criminal threads in one place before deciding how to update your own witness-handling process.
Read or listen: Legal Futures
Practice Prompt
Try the below prompt to build a structured schedule for gathering a client's or witness's own account and instructions, so their evidence is captured in a controlled, documented way rather than shaped by an unsupervised chatbot session. It draws on the approach behind our new Client Instruction Schedule tool, and it fits this week's theme of keeping a witness's account authentic. 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 litigation solicitor in England and Wales. Your task is to design a structured schedule for gathering a client's or witness's own account and instructions on a matter, issue by issue, as a planning aid only. The aim is a clear, lawyer-controlled record of what the person actually recalls and instructs, so their evidence stays authentic and any AI use is surfaced rather than hidden.
Context to apply:
- The matter: {claim type, court or track, and the main issues in dispute}
- The person: {e.g., "the client, who is also a witness of fact" / "a third-party witness"}
- What you already have: {e.g., "a short instruction email and some documents"}
- Known sensitivities: {e.g., "vulnerable witness", "reporting restrictions", "litigant has been using AI tools"}
Produce:
1. An issue-by-issue schedule
For each issue in dispute, set out the specific questions to put to the person to capture their own recollection and instructions in their own words, kept neutral and non-leading so you are not suggesting the answer.
2. An AI-use check
Include a short set of questions that establish whether the person has used any AI tool to prepare, summarise or rehearse their account, what they entered, and what the tool produced, so any such use can be recorded and considered for disclosure.
3. A documents and gaps list
For each issue, note the documents that should corroborate the account and flag where the instructions are thin, inconsistent, or need a follow-up.
4. A record and next steps note
Set out how to record the person's answers and any AI use on the file, and the next steps before a formal statement is drafted.
Constraints:
- {Add matter-specific constraints, for example the relevant pre-action protocol, any reporting restriction, or a vulnerable-witness consideration.}
- Apply the law and procedure of England and Wales only.
- Do not draft the witness statement itself, and do not put words in the witness's mouth: the output is a question set and a record, not evidence.
- Do not invent facts, cases, or rules. Where something is unclear, flag it as a question.
- This is a planning aid, not legal advice, and the supervising solicitor remains responsible for the evidence and for compliance with the rules on witness statements.How did we do?
Hit reply and tell me what you would like covered in future issues or any feedback. We read every email!
Thanks for reading,
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!




