Yes, AI-assisted expert evidence can survive cross-examination, but only where the opinion remains the expert’s own, the use of AI has been appropriately disclosed, and every step of the analysis can be explained, verified and reproduced. AI-generated opinion, by contrast, will not survive.
The gap between those two positions is now the most important conversation for a solicitor to have with an expert before instruction.
Three developments have sharpened the question in the space of fifteen months: pointed judicial comments from Mr Justice Waksman in November 2025, the updated judicial guidance on artificial intelligence published on 31 October 2025, and the Academy of Experts guidance for expert witnesses published on 30 January 2026. This article looks at each of them, and at what a forensic accountant’s evidence needs to look like to withstand challenge.
What does CPR Part 35 require of expert evidence?
Nothing in CPR Part 35 prohibits an expert from using AI. What it does require is unchanged: under rule 35.3, the expert’s duty is to help the court on matters within their expertise, and that duty overrides any obligation to the instructing party. The report must comply with Practice Direction 35, set out the substance of all material instructions, and end with a statement that the expert understands and has complied with their duty to the court, verified by a statement of truth.
Those requirements are where AI-assisted evidence is tested. An expert who cannot explain how a figure was produced, because a generative tool produced it, cannot honestly sign the statement of truth. An expert whose opinion was shaped by an AI-generated draft has a problem with independence. The rules did not need amending to deal with AI; they already ask the questions that expose its misuse.
What did Mr Justice Waksman say about AI and expert reports?
Speaking at the Bond Solon Expert Witness Conference on 7 November 2025, Mr Justice Waksman, head of the Technology and Construction Court, addressed a striking finding from that year’s expert witness survey: an expert had been asked to accept an instruction in which the solicitor insisted on supplying an AI-generated draft of the expert report. He described that conduct as a “gross breach of duty” on the part of the solicitor.
He was equally troubled that 14 per cent of surveyed experts said they would accept an instruction on that basis, taking the view that this was inappropriate even where the expert intended to discard the draft. His broader message to experts was to avoid using AI to answer the very questions they have been instructed to answer, because doing so compromises the independence on which their evidence depends.
For solicitors, the practical point cuts both ways. It is not only the expert’s use of AI that can undermine evidence. An instruction that arrives with a machine-drafted report attached exposes the instructing solicitor to criticism, and taints the expert’s opinion before a word of it has been written.
What does the October 2025 judicial guidance say?
The updated judicial guidance on AI was published on 31 October 2025, replacing the April 2025 version. It is addressed to judicial office holders rather than to experts, but it reveals exactly how the judiciary now thinks about these tools, and that thinking will follow an expert into the witness box.
Key themes include:
- Hallucination: the guidance expands its treatment of AI-generated output that is incorrect or misleading, including fabricated citations and invented facts.
- Confidentiality: information entered into a public AI chatbot should be treated as if it were published to the world. For an expert handling disclosure, medical records or financial data, that principle is decisive against public tools.
- Accountability: Lord Justice Birss, the lead judge for AI, emphasised that judicial office holders remain personally responsible for everything produced in their name. The same logic applies with full force to an expert’s report.
- Manipulation risks: the guidance now warns about hidden “white text” designed to be read by machines but invisible to human readers, a reminder that AI-processed documents can be gamed.
A judge reading expert evidence in 2026 has been trained to ask where AI might have introduced error, and who checked. Experts should expect cross-examining counsel to ask the same.
What does the Academy of Experts January 2026 guidance require?
The Academy of Experts guidance on the use of AI, published on 30 January 2026 with a foreword by Lord Neuberger, is the first dedicated guidance for expert witnesses. It sets out background on AI and its risks, guidance on compliant use, and a practical checklist.
Its most useful contribution is the distinction between high-risk and low-risk uses:
- High-risk uses include generating substantive content or carrying out material analysis on which the expert’s opinion will be based. These should be disclosed to instructing lawyers, and the expert should ensure there are no objections before proceeding.
- Low-risk uses include summarising or extracting data on which the expert will not rely. These may not require disclosure, but the expert must still be prepared to explain their use of AI under cross-examination.
- In cases of doubt about where a use falls, the guidance points towards disclosure.
The message for experts is that AI can assist with analysis, organisation and quality control, but it cannot replace expert judgement, and the expert remains responsible for compliance with their duties to the court at all times. Experts are also reminded to check whether a tool uses AI at all, which is increasingly hard to tell as AI is embedded into everyday research and analysis software.
What will cross-examination on AI use look like?
Counsel challenging AI-assisted evidence will not need new techniques. The questions are familiar ones about methodology, asked with a new target:
- Did you use AI at any stage of preparing this report, and which tools?
- What did the tool do, and what did you do?
- How did you verify the output, and against what source material?
- Could you reproduce this analysis and reach the same figures?
- Was confidential case material entered into any public AI system?
- Is the opinion in this report yours, or the tool’s?
An expert with a documented methodology answers those questions comfortably. An expert who cannot say what the tool did, or who treated its output as the answer rather than as material to be tested, will watch their evidence unravel.
How does AI-assisted expert evidence survive challenges?
In our experience, five conditions separate evidence that survives from evidence that does not:
- Ownership of the opinion. Every conclusion is the expert’s, reached by applying their own judgement to verified data. AI accelerates the work; it does not perform the reasoning.
- Appropriate disclosure. Material AI use is raised with instructing solicitors at the outset and, where appropriate, explained in the report itself, consistent with the Academy of Experts guidance.
- Verification. Every figure traces back to source records, and the checking process is documented, not asserted.
- Reproducibility. The analysis can be re-run and explained step by step, which is also what allows the opposing expert to test it.
- Data security. Case material is processed only in secure, controlled environments, never in public chatbots.
Handled this way, AI strengthens rather than weakens expert evidence, because it allows the whole dataset to be analysed instead of a sample, at a proportionate cost. The question in the witness box then becomes one the expert wants to answer.
Frequently asked questions
Do expert witnesses have to disclose their use of AI?
Under the Academy of Experts guidance published in January 2026, high-risk uses, such as generating substantive content or material analysis underpinning the opinion, should be disclosed to instructing lawyers before proceeding. Low-risk uses may not need disclosure, but the expert must be prepared to explain any AI use under cross-examination.
Can a solicitor give an expert an AI-generated draft report?
No. Mr Justice Waksman described a solicitor’s insistence on supplying an AI-generated draft expert report as a gross breach of duty, in comments made at the Bond Solon Expert Witness Conference in November 2025. He also criticised experts who would accept instructions on that basis, even if they intended to discard the draft.
Does CPR Part 35 ban the use of AI in expert reports?
No. CPR Part 35 does not mention AI. It requires the expert’s overriding duty to the court, a report compliant with Practice Direction 35, and a statement of truth. AI use becomes a problem where it prevents the expert from honestly meeting those requirements, for example where the expert cannot verify or explain the output.
What happens if AI invents a citation or figure in an expert report?
The consequences can be severe. In Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), the High Court warned that generative AI tools cannot be relied on for legal research and that putting fabricated authorities before the court can attract serious sanctions, potentially including criminal liability.
Will AI use make expert evidence inadmissible?
Not in itself. Admissibility turns on the established Part 35 framework: relevance, expertise, independence and compliance with the rules. Responsible, disclosed, verified AI assistance does not offend any of those requirements. Undisclosed or unverified AI use, or AI-generated opinion, goes to the weight and credibility of the evidence and can destroy both.
Talk to us before you instruct
Frenkels Forensics has over 40 years of experience providing expert forensic accountancy evidence, and our suitably qualified experts regularly give evidence under cross-examination. Whether your matter involves crime and fraud, commercial litigation or valuations, or personal injury/clinical negligence personal injury, we are happy to explain exactly how we use technology, and how we do not, before you instruct.
Call us on 0330 118 8200 or email help@frenkels.com.