The Court of Appeal has recently held that a witness’s use of AI to prepare for cross examination amounted to prohibited witness coaching. The decision has important implications for anyone involved in litigation – whether as a witness, a party or a legal advisor.

The Decision

In R v FGD [2026] EWCA Crim 918, the Court of Appeal (Criminal Division) considered a prosecution appeal against the trial judge’s decision to stay proceedings as an abuse of process after it emerged that the sole prosecution witness had used a generative AI tool to prepare for cross examination.

The case concerned an allegation of rape. During the trial, a phone download revealed two documents on the complainant’s device containing a summary of their account of events followed by a series of questions “in the form of a cross examination” together with suggested answers. When asked who they had been communicating with, the complainant responded: “I had put stuff into AI to help me prepare myself for court” (see paragraph 5).

The AI generated material was extensive.

Paragraph 5 recorded content from the introductory section from the AI, which read: “what I am going to do is pull together everything you’ve shared about that night into one clear picture, so that you can see how strong and consistent your account actually is. Then I’ll go through every question a defense lawyer might ask with suggested answers.” What followed was several pages of text containing 24 questions and suggested answers on various topics. The trial judge concluded this amounted to coaching and stayed the proceedings, finding that a fair trial was impossible.

The Court of Appeal allowed the appeal. Lady Justice May held that the Recorder had erred in concluding that the trial process could not deal with the prejudice caused by the  complainant’s use of AI. The Court notes that the complainant’s evidence in chief had been given by pre-recorded interview some years before and was therefore uncontaminated, and the complainant’s answers in cross examination could be assessed against the coaching content of AI.

Importantly, at paragraph 18, the Court also observed that, unlike traditional coaching conducted behind closed doors and not fully recorded, the AI-generated material was fully documented and could be placed in its entirety in front of the jury.

However, the Court issued a broader warning that went well beyond the facts of the individual case at paragraph 23:

“the coaching of witnesses in their evidence, by any means, is not permitted. All witnesses, whether for the prosecution or defence, should be firmly discouraged from resorting to the use of AI to prepare for giving evidence. Doing so could run the risk of unfairness up to the point, in an extreme case, where the court may have to consider excluding the evidence.”

The Court further warned in paragraph 23 that the use of AI by witnesses “may open up further lines of enquiry, exposing their online activity to welcome, wide ranging and potentially more intrusive investigation and examination for the purpose of disclosure, than would otherwise have been required” and that it may put “personal information into the public domain, in open AI, in a way that may never have been anticipated.”

Witness Familiarisation versus Witness Coaching: A Critical Distinction

The distinction between permissible witness familiarisation and impermissible witness coaching is well established. In R v Momodou [2005] EWCA 177 the Court of Appeal confirmed that familiarisation, helping a witness understand the court process and what to expect when giving evidence, is entirely proper. Coaching, rehearsing or shaping the substance of a witness’s evidence, is not.

The 2026 decision applies this framework to AI. When a witness uses an AI tool to generate anticipated questions and model answers, the effect is the same as a third party coaching the witness on what to say. The witness is exposed to suggested formulations that do not originate or come from their own memory, and there is a real risk that those suggestions will influence the evidence they ultimately give.

The Wider Regulatory Landscape

This decision sits within a broader regulatory context. The judiciary and Civil Justice Council have been considering the implications for AI in the context of court proceedings for some time.

The Civil Justice Council’s working group on the use of AI published an interim report in early 2026. One of the principal areas of concern was the drafting of witness statements. The CJC notes that responses reflected differing views on whether additional safeguards or disclosure requirements are needed for witness evidence and noted the overarching concern was to preserve the “authenticity, integrity and personal recollection of the witness” and concern that AI tools could “reshape, embellish or otherwise influence evidence in ways that may not be immediately apparent.”

The SRA has also signalled its concern. Solicitors owe professional obligations under the SRA Code of Conduct to act with integrity (Principal 5), to act in the best interests of their client (Principle 7) and not to mislead the court (paragraph 1.4 of the Code of Conduct for Solicitors). The SRA’s broader messaging on technology use makes clear that these obligations apply with full force to AI: a solicitor who allows or fails to prevent a witness from using AI to shape their evidence risks breaching their duty to mislead and their obligation to uphold the proper administration of justice. Where AI use comes to light after a statement has been served, questions of professional conduct may follow.

Practice Direction 57AC then governs trial witness statements in the Business and Property Courts and requires that statements are to be in the witnesses “own language” and set out only what the witness “claims personally to recollect.” The use of AI to generate or rehearse answers sits uncomfortably alongside those requirements and reinforces the conclusion reached by the Court of Appeal.

Practical Guidance for Witnesses and their Legal Teams

The practical position is clear. Witnesses must not use Chat GPT, Copilot, Claude, Harvey or any generative AI tool, whether private or public, to:

• Generate possible cross examination questions;
• Suggest answers to questions that may be asked;
• Test or assess the strength of their own evidence;
• Explain or rationalise inconsistencies in their evidence; or
• Otherwise assist with the consideration or preparation of witness evidence.

Evidence must always be based on the witness’s own recollection and understanding of the relevant events. The use of AI crosses the line from familiarisation into coaching because it introduces external suggestions about the substance of what a witness will say.

Disclosure Risk

There is a separate and important practical concern about disclosure. Any interactions with AI tools, including prompts entered and responses generated, may be recorded. Those records could potentially become disclosable in litigation, whether under standard disclosure or a specific disclosure application.

The consequences of this are significant. If a witness has used AI to prepare for cross examination, the opposing party may be entitled to see the full record of what was asked and what AI suggested. That material could then be deployed in cross examination to undermine the witness’s credibility – putting to them that their evidence reflects AI generated suggestions rather than a genuine recollection. Even where the witness insists they were not influenced, the existence of an AI interaction creates a vulnerability that is difficult to overcome.

What Remains Permissible

It is worth emphasising what witnesses can still do. Witness familiarisation, understanding the hearing process, what to expect when giving evidence and the general guidance on how to give clear answers remains entirely proper. Discussing factual matters with legal representatives in the ordinary course of proof taking is also unaffected provided those discussions are directed at eliciting the witness’s own recollection rather than scripting answers.

The key question is whether the activity shapes the substance of the witness’s evidence, or helps them to understand the process of giving it.

Conclusion

Proactive instruction to witnesses on AI use is essential. It can no longer be said that witnesses will not think to use these tools to assist their evidence and drafting. Legal teams should incorporate clear guidance on AI use into their witness preparation protocols as a matter of course, to protect the integrity of the evidence and the witnesses from the consequences of noncompliance.

For more information about the content of this article, please contact Andy McGregor.