AI and witness evidence: navigating the risks

Over the last decade or so, there has been increasing focus on the interaction between the limits of human memory and the role of judges in resolving factual disputes. In 2013, Lord Leggatt (as a High Court Judge) called attention to scientific research on the fallibility of human memory. He argued that, in commercial cases, judges should “place little if any reliance at all on witnesses' recollections … and to base factual findings on inferences drawn from the documentary evidence and known or probable facts” Gestmin SGPS SA v Credit Suisse (UK) Ltd & Anor [2013] EWHC 3560 (Comm).

In 2021, Practice Direction 57AC was introduced in the Business and Property Courts, which was intended to address these kinds of concerns and to enhance the relevance and reliability of witness evidence through additional rules on the preparation of witness statements.

Fast forward to today, when the use of generative AI (Gen AI) tools is becoming increasingly ubiquitous in litigation and legal practice generally. This raises questions for solicitors: what different angles does Gen AI offer to this debate? Can Gen AI help make witnesses’ accounts more relevant and reliable or does it risk having the opposite effect?

These questions have not been answered definitively. However, at present, our view is that the answer emerging from recent cases and guidance is that both solicitors and the witnesses they advise should proceed with caution when considering the use of AI in the context of witness evidence.

Witness statements: Godwin v Godwin and the Civil Justice Council consultation

In Godwin v Godwin [2026] EWHC 923 (Ch), a judgment released in April 2026, the High Court considered the use of AI tools (specifically ChatGPT, i.e. a large language model) to assist in the preparation of witness statements. The case related to probate but has implications for any cases in the English courts.

In this case, in which the parties were unrepresented, two witnesses (one of whom was the defendant) stated in their witness statements that they had “limited assistance” from a “digital assistant” (which was ChatGPT) for grammar, spelling and “presentation”.

Although this may, on the face of it, seem like a relatively unsurprising use of AI, the judge was concerned with the impact on reliability and compliance with the CPR, holding that both witnesses were “sophisticated people” and “there was no good reason…to use ChatGPT in this case”, particularly as Microsoft Word has grammar and spelling checking functionality. He concluded ultimately that he could not be sure that their witness statements were in their own words in accordance with the requirements in the CPR and must treat their witness statements with caution.

In 2025, the Civil Justice Council (CJC) started a consultation on the use of AI in court documents, chaired by the Chancellor of the High Court. In June 2026, the CJC provided an update on its findings from the consultation. It noted that there was a consensus amongst respondents that for documents such as pleadings, no additional formal requirements relating to AI are currently necessary. However, it identified that the use of AI in preparing witness statements was “the principal area of ongoing consideration” and noted concerns “about the potential for AI tools to reshape, embellish or otherwise influence evidence in ways that may not be immediately apparent” – but also noted that such issues are not novel issues created by AI. Amongst the respondents raising concerns was the Law Society, which, in its response, called for all witness statements to contain a declaration as to non-AI use. The CJC’s full report is expected at the end of the year.

In August 2026, the SRA issued a warning notice regarding the use of AI, including highlighting the risk of hallucinations impacting the accuracy of court documents. While the examples in the warning notice mainly concerned legal submissions, the SRA’s warnings in relation to professional duties concerning materials put before the courts of course also apply to evidence.

Preparing to give evidence: R v FGD

In R v FGD [2026] EWCA Crim 918, released in June 2026, the Court of Appeal (Criminal Division) addressed the use of AI by witnesses to prepare for cross-examination. While it concerned a criminal trial, the lessons have implications for both civil and criminal cases.

During cross-examination in the proceedings at first instance, counsel for the defence requested a full download of a witness’ phone. Documents were subsequently produced showing a series of questions and suggested answers. Upon being questioned, the witness confirmed that “I had put stuff into AI to help me prepare myself for the court”. The documents contained statements such as: “What I am going to do first is pull together everything you’ve shared about that night into one clear picture… then I’ll go through every question a defence lawyer might ask with suggested answers.” The defence applied to stay the proceedings on the grounds of abuse of process. The first instance judge was satisfied that the documents were evidence of witness coaching (regardless of whether they were prepared by AI or a person), and granted the application. The prosecution appealed.

The Court of Appeal confirmed that “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 evidence.”

However, the Court of Appeal did ultimately allow the appeal on the basis that this matter did not prevent a fair trial and ordered that the trial would proceed. Interestingly, the considerations included that the witness’s evidence in chief had been uncontaminated by AI (because it was a police station interview recorded soon after the criminal complaint was made) and that the apparent AI cross-examination coaching could be disclosed: “all that the AI “coach” provided to [the witness] was fully recorded and could be placed in its entirety before the jury”. This highlights another risk of witnesses using AI for preparing for cross-examination, namely disclosure risks. The comment about having assurance that the witness’ evidence in chief was uncontaminated due to it being a recording would of course not apply to civil proceedings.

Conclusion

The use of AI in litigation and legal practice generally is likely to continue to grow. Ensuring responsible AI use by solicitors will remain a top priority for the judiciary and regulators.

Crucially, as we have outlined, the debate around the use of AI in witness evidence has another dimension. It intersects with another topic which has long been an area of concern for the judiciary: the reliability and relevance of witness evidence and adherence to requirements intended to manage these risks, such as in PD57AC. Practitioners must also note that failures to comply will present significant risks to the outcome of litigation, such as reduced weight being placed on witness evidence, disclosure risks and adverse judicial comments on the conduct of a case.

While additional case authority and guidance will emerge, including potential rule changes following the CJC’s final report, practitioners must bear in mind that the interaction between Gen AI use and witnesses needs particular care. This applies both to how solicitors prepare, or supervise the preparation of, witness statements and to how they provide appropriate guidance to witnesses regarding their own use of AI.

Finally, practitioners should take a risk-based approach. There are several Gen AI use cases relevant to witness evidence, but which do not present the same types of risks to the reliability of witness evidence put before the court. These would include the use of Gen AI to assist in the creation of a chronology of documents to discuss with a witness, or to identify potential inconsistencies with a witness statement submitted by the other side.