Witnesses using AI to prepare for court has stopped being a theoretical worry. In R v FGD the Court of Appeal (Criminal Division) dealt with a rape trial that collapsed at first instance because the complainant had asked a chatbot to get them ready for cross-examination. The judgment was handed down on 4 June 2026 and has drawn attention this month. It carries a plain message that firms need to pass on to clients and to every witness from whom they take a statement.
What happened at trial
The complainant was the only prosecution witness. Under cross-examination they were challenged about where they had been at a particular point that evening. Going home that night, they searched their cloud storage and produced screenshots to answer the point. When the prosecution asked to recall them, the defence sought a full download of the phone, and the recorder ordered it. The download revealed two documents. One set out a summary of what the complainant recalled of the night in question. The other ran to several pages of questions in the form of a cross-examination, twenty-four of them, with suggested answers grouped under headings such as consent and capacity. Asked who they had been communicating with, the complainant answered that they had put material into AI to help them prepare for court. The recorder held that this amounted to witness coaching, which has been forbidden since Momodou, and stayed the proceedings as an abuse on the footing that a fair trial had become impossible.
Why the appeal succeeded
Lady Justice May, sitting with Mr Justice Griffiths and His Honour Judge Picton, allowed the prosecution appeal under section 58 of the Criminal Justice Act 2003 and sent the case back to be relisted. The court did not doubt that the documents were a form of forbidden coaching. Its objection was that a stay is an exceptional measure of last resort, and nothing in the ruling showed which remedies the recorder had weighed before concluding that no fair trial was possible. Two features of the case told against a stay. The complainant's evidence in chief was a recorded interview given years earlier, long before any chatbot came into it, so only the answers in cross-examination were open to challenge. And unlike coaching conducted behind a closed door, everything the machine had said was recorded in full, so it can be put before a jury or reduced to agreed facts.
The warning to witnesses
The court closed with a passage aimed beyond this case. All witnesses, for the prosecution or the defence, "should be firmly discouraged from resorting to the use of AI to prepare for giving evidence". Doing so, the court said, "could run the risk of unfairness up to the point, in an extreme case, where the court may have to consider excluding evidence". A second risk sits alongside the first and gets less attention. For complainants in cases involving allegations of sexual misconduct, turning to AI opens fresh lines of enquiry, exposing online activity to a wider and more intrusive examination for disclosure than would otherwise have been needed. It also puts personal information into the public domain in a way the witness never intended. That is what happened here, where an entire phone was downloaded. The court noted its own concern at how broad that order was.
What your firm should do
The practical step is small and belongs in your standard papers. When you send a witness their statement to approve, or write to a client about a hearing, add a line telling them not to ask any AI tool to help them prepare their evidence, and give the reason. It reads as coaching, it invites a search of their device, and it puts their own account in question. Witness familiarisation is still permitted and remains worth doing properly. Coaching is not, and the line does not shift because software rather than a person is doing it. Record that you gave the warning, so the file shows it. The court suggested that the National Police Chiefs Council, the Crown Prosecution Service and the Criminal Procedure Rules Committee consider how to regulate AI in criminal proceedings, and observed that this case is unlikely to be an isolated example. Firms that get the warning into their client care wording now will not be rewriting it under pressure later.
The judgment is published in full as R v FGD [2026] EWCA Crim 918 on the National Archives case law service.
If your client care letters and witness packs have never been read with AI in mind, that is a short, contained piece of work and a sensible place to start: talk it through with us.