Since 2023, courts in England and Wales have been issued judicial guidance requiring expert witnesses to disclose whether and how AI tools were used in preparing their reports. A forensic accountant who uses AI to assist analysis without disclosure risks their report being challenged, given reduced weight, or excluded. The obligation covers any AI tool that contributed to the analysis, drafting, or conclusions of the expert witness report.
- Why are AI disclosures now required in expert witness reports?
- What do the current CJC guidance and PD57AC say about AI use by expert witnesses?
- What we see in practice: how forensic accountants are and are not using AI tools
- What must a forensic accountant disclose about AI use in their report?
- What risks does undisclosed AI use create in financial proceedings?
- How should solicitors ask about AI use when instructing a forensic accountant?
Why are AI disclosures now required in expert witness reports?
AI disclosures are required in expert witness reports because the personal duty of the expert to the court extends to the methods used to form their opinion. CPR Part 35 and Practice Direction 35 require the expert to certify that the report contains their true and complete professional opinion. Where an AI tool has contributed to forming or expressing that opinion, the court and the opposing party are entitled to know what tool was used, how it was used, and what verification the expert applied to its output.
The requirement emerged from two converging developments. First, the rapid adoption of generative AI tools (including large language model systems) by professionals in many fields from 2023 onwards, which created a genuine risk that expert reports would be drafted with AI assistance in ways that were not apparent from the face of the document. Second, a series of high-profile incidents in courts in the United States and, to a lesser extent, in England and Wales, where AI-generated content was included in legal documents without verification, producing fabricated citations and erroneous analysis. Courts became alert to the possibility that expert reports, like other professional legal documents, could contain AI-generated content that the expert had not adequately verified.
The principle underlying the disclosure requirement is consistent with the core obligation of the expert witness: the expert must be able to stand behind every conclusion in the report. If an AI tool produced an analysis and the expert adopted it without understanding the methodology the tool applied, the expert cannot genuinely certify that the report contains their reasoned professional opinion. Courts require disclosure so that they can assess whether the expert's verification process was adequate and whether the AI-assisted conclusions should be treated as reliably expert opinion or given reduced weight.
In England and Wales, judicial guidance on AI use in legal proceedings has developed incrementally since 2023. The Civil Justice Council published interim guidance in 2023 and updated guidance in 2024, and the Senior Courts judiciary have issued guidance to specific court divisions. The direction of travel is clear: disclosure is expected, non-disclosure creates a procedural risk, and expert witnesses who use AI tools without appropriate verification and disclosure are exposed to challenge on the integrity of their evidence. For the broader context of what an expert witness report must contain, see our article on forensic accounting expert report contents.
What do the current CJC guidance and PD57AC say about AI use by expert witnesses?
The Civil Justice Council interim guidance on AI (published 2023, updated 2024) addresses AI use in civil proceedings broadly and applies to expert witnesses as a specific category of legal professional whose obligations under CPR Part 35 interact with the new AI disclosure requirements. The guidance does not prohibit AI use by expert witnesses. It requires transparency about AI use that is material to the formation of the expert's opinion.
The CJC guidance distinguishes between AI tools used for administrative purposes (drafting letters, formatting documents, summarising disclosed materials) and AI tools used in ways that contribute to the substance of the expert's analysis or conclusions. The disclosure obligation is triggered by the second category. An expert who uses a standard word processor with AI-assisted grammar checking is not required to disclose that use. An expert who uses a large language model to analyse a set of financial transactions, to identify patterns in accounting data, or to draft sections of the analytical narrative in the report, is using AI in a way that is material to the substance of the report and must disclose it.
Practice Direction 57AC, which governs witness statements in Business and Property Courts proceedings, contains an analogous requirement that the witness confirm they have not received assistance with the drafting of their statement beyond what is proper. While PD57AC does not govern expert reports directly (expert reports are governed by CPR Part 35 and PD35), the disclosure philosophy is the same: the court and the opposing party are entitled to know what assistance was used in producing the document, so that the weight to be given to it can be properly assessed.
As at 2026, there is no single rule in CPR Part 35 or PD35 that sets out an explicit AI disclosure checklist for expert witnesses. The disclosure obligation is instead derived from the existing requirements: the expert's duty of candour to the court, the requirement that the report represent the expert's own reasoned opinion, and the requirement to disclose the basis on which that opinion was formed. Courts have made clear in case management decisions from 2024 and 2025 that they expect experts to address AI use in the declaration section of their reports, and some court directions now explicitly require experts to address the AI question in their written evidence.
The Law Commission and the Civil Justice Council are each engaged in ongoing work on AI and the courts as at 2026. The direction of travel is towards more specific guidance, and solicitors should expect more detailed rules to emerge from the CJC's full AI report, expected in 2026. For now, the safest and most professionally appropriate course for any forensic accountant is to disclose all material AI use in the declaration section of every report.
What we see in practice: how forensic accountants are and are not using AI tools
From forensic accounting practice in England and Wales across financial remedy, commercial litigation, HMRC investigations, and POCA confiscation proceedings, I can identify a clear pattern in how AI tools are currently being used by forensic accountants, and where the risks of challenge and non-disclosure are most acute.
There are forensic accounting tasks where AI tools provide genuine, verifiable assistance with relatively limited risk. Transcribing handwritten or scanned financial records into structured data is one. AI-assisted optical character recognition and data extraction from PDFs and bank statements has improved substantially since 2023 and reduces the manual data entry burden in large volume cases. The output of that process is a dataset of financial transactions that the forensic accountant then analyses using their professional judgement. Used in this way, AI is a tool for data capture, not for analysis, and the disclosure requirement is manageable: the forensic accountant states in the report which tool was used for data extraction and that all extracted data was verified against source documents.
Summarising long disclosure bundles is another area where AI tools are used. A large commercial litigation case may involve tens of thousands of pages of financial disclosure. AI tools can identify which documents are likely to be relevant to specific issues, allowing the forensic accountant to focus their review on the most material documents. This is a document triage function, not an analytical function, and the forensic accountant still reads and analyses the relevant documents themselves. Disclosure of this use is appropriate and straightforward: the expert notes in the documents reviewed section that AI-assisted document triage was used to identify relevant materials from the disclosure set, and that the conclusions of the report are based on the expert's own review of the documents so identified.
The area of genuine risk and challenge is where AI tools are used to draft analytical sections of the expert witness report, to formulate the expert's opinion on specific issues, or to perform the financial calculations that underpin the conclusions. I am aware of instances in 2024 and 2025 where forensic accounting experts used large language model tools to draft sections of reports addressing complex accounting issues and served those reports without disclosure of the AI assistance. In each case identified, the challenge arose not at the point of service but at the joint meeting or at trial, where cross-examination on the analytical reasoning exposed that the expert could not fully explain the basis of conclusions that had in substance been generated by an AI tool rather than by the expert's own reasoning.
The fundamental problem with using AI to perform or draft the substantive analysis in a forensic accounting expert report is that AI tools cannot be cross-examined, do not hold professional qualifications, and do not owe a duty to the court. The expert who signs the declaration under PD35 certifies that the report contains their reasoned professional opinion. If a material part of that opinion was formulated by an AI tool, the certification is misleading. Courts have treated this seriously in the post-2023 cases where it has arisen, and the consequences for the expert's professional credibility have been significant.
AI tools currently available to forensic accountants, including specialist legal and financial AI platforms as well as general large language models, have a specific limitation that is critical in the forensic context: they can generate plausible-sounding analysis that is factually incorrect or that applies an incorrect methodology to the specific facts of the case. In forensic accounting, where the figures must be traceable to source documents and the methodology must be defensible at trial, this limitation is disqualifying for any AI output that is not independently verified line by line by the expert. The verification process required to use AI-generated analysis reliably in a forensic report is often more time-consuming than simply performing the analysis directly. For straightforward tasks like data extraction, AI tools save time. For complex analytical tasks, the verification burden often outweighs the drafting saving.
What must a forensic accountant disclose about AI use in their report?
A forensic accountant using AI tools in preparing an expert witness report should disclose in the body or declaration of the report: the specific tool or tools used; the function for which each tool was used; the verification process applied to the AI output; and a statement of whether the conclusions of the report are the expert's own reasoned opinion or are in any way the output of an AI tool.
The disclosure should be specific rather than generic. A statement that AI tools were used in the preparation of this report is inadequate: it does not tell the court or the opposing party what was done, how, or what verification was applied. A specific disclosure might read: optical character recognition and data extraction from the 47 bank statements disclosed were assisted by [tool name]. All extracted transaction data was verified against the source documents before being used in the analysis. No AI tool was used to perform any financial calculation or to draft any section of the analytical narrative in this report. The conclusions expressed in this report are the expert's own reasoned professional opinion based on the documents reviewed and the calculations performed by the expert personally.
Where an AI tool was used to draft any part of the analytical narrative, the disclosure must say so and must explain what verification the expert applied to the AI-drafted content. A forensic accountant who instructs an AI tool to draft a section on, for example, the accounting treatment of specific inter-company transactions, and then reviews and adopts that section without modification, has not applied adequate verification: the expert must be able to explain the reasoning in that section in their own words. Where the expert has reviewed, tested, and substantially revised AI-drafted content, disclosure of the original AI assistance and the expert's verification process is sufficient and appropriate.
In HMRC investigation and POCA confiscation cases, see our service pages on HMRC investigations and POCA confiscation for the specific context of expert witness reports in those proceedings. In matrimonial finance cases involving business valuation or asset reconstruction, the same disclosure principles apply and the same verification standards are required. See our service page on matrimonial finance for the context of forensic accounting instructions in those proceedings.
What risks does undisclosed AI use create in financial proceedings?
Undisclosed AI use in a forensic accounting expert report creates risks at three levels: the procedural, the substantive, and the professional.
The procedural risk is that the opposing party or the court discovers the undisclosed AI use during the proceedings, through cross-examination of the expert on their methodology, through the joint meeting process, or through questions under CPR 35.6. Where AI use is discovered and has not been disclosed, the court may treat the non-disclosure as a breach of the expert's duty of candour. The consequences can include reduced weight given to the report, exclusion of specific AI-generated conclusions, or in serious cases exclusion of the report entirely. Courts in England and Wales have not yet excluded a forensic accounting report solely on the grounds of undisclosed AI use as at 2026, but judicial guidance has been emphatic that disclosure is expected and that non-disclosure creates a serious procedural risk.
The substantive risk is that undisclosed AI-generated analysis contains errors that are not identified until trial. Large language model tools operating on financial data can produce analysis that appears internally consistent but is methodologically incorrect for the specific facts of the case. A forensic accountant who serves AI-generated analysis without adequate verification may find at trial that the cross-examining counsel, or the opposing expert in the joint meeting, has identified an error in the AI output that the expert cannot explain and did not catch. The expert's credibility on all issues in the report is damaged, not just the specific issue on which the error was found.
The professional risk is the most long-lasting. A forensic accountant whose report is found to contain undisclosed AI-generated analysis faces potential referral to their professional body. The ACCA and ICAEW each have conduct standards requiring members to be honest about the basis of their professional work and to not mislead clients, courts, or third parties. An expert who certifies under PD35 that the report contains their true and complete professional opinion when material sections were generated by an AI tool without adequate verification is potentially in breach of those professional conduct standards, regardless of whether the courts takes formal action in the proceedings.
For solicitors, the risk of undisclosed AI use by an instructed expert extends to their own exposure to wasted costs applications if the report is challenged and the proceedings are delayed as a result. Solicitors should ask directly about AI use when instructing forensic accountants and should include AI disclosure as a standard term in their letters of instruction.
How should solicitors ask about AI use when instructing a forensic accountant?
Solicitors should address AI use explicitly in the letter of instruction to the forensic accountant, and should follow up on the answer before the report is served. The letter of instruction should include a specific question or requirement on the following lines: please confirm in your report whether any AI tools were used in the preparation of this report, specifying the tool, the function for which it was used, and the verification process applied to its output. Please also confirm that the opinions expressed in the report are your own reasoned professional judgement and not the output of an AI tool.
When reviewing a draft report before service, the solicitor should check whether the report contains an AI disclosure statement and whether that statement is specific enough to satisfy the court's requirements. A generic disclaimer that AI may have been used in preparation is not adequate: the statement must identify what was used and for what purpose. Where the draft report contains no AI disclosure, the solicitor should ask the expert directly whether any AI tools were used and, if so, whether they should be disclosed.
At the point of instructing a new forensic accountant, solicitors should include AI use as a standard topic in their initial discussion alongside qualifications, experience, and availability. The relevant questions are: do you use AI tools in your forensic accounting work; if so, which tools and for what purposes; what verification process do you apply to AI-generated output; and how do you disclose AI use in your expert reports. An expert who is evasive or who cannot give specific answers to these questions is a higher-risk instruction than one who can explain clearly both what they use AI for and where they rely entirely on their own professional judgement.
A forensic accountant who uses AI tools appropriately, with clear verification processes and transparent disclosure, is not at a disadvantage compared to one who uses no AI. The disclosure, properly handled, is a mark of professional transparency. An expert who can say clearly in their report that AI was used for data extraction from 2,400 bank statement pages, that the output was verified against source documents, and that the analysis, calculations, and conclusions were performed by the expert personally, is in a stronger position than one who either uses AI without disclosure or refuses to use any tool that might assist efficiency. The question is not whether AI use is permitted: it is whether the use is disclosed, verified, and within the scope of tools that assist rather than replace the expert's own analysis.
Frequently asked questions
Is a forensic accountant allowed to use AI at all?
Yes. No current rule in CPR Part 35, PD35, or the CJC guidance prohibits forensic accountants from using AI tools. The requirement is transparency and verification, not prohibition. AI tools used for data capture, document triage, or formatting assistance do not trigger the same disclosure obligation as AI tools used to perform analysis or draft substantive sections of the report. The critical test is whether the AI tool contributed to the expert's opinion on any issue: if it did, that use must be disclosed with specifics about the tool, the function, and the verification applied.
What must be disclosed if AI is used to analyse financial data?
Where AI is used to analyse financial data, the disclosure must identify the specific tool used, describe the type of analysis performed, explain what inputs were provided to the AI tool, and set out the verification process the expert applied to the output. If the AI output was used directly in the report without modification, that must be stated clearly. If the AI output was used as a starting point and substantially revised by the expert applying their own professional judgement, the disclosure should explain the extent of the expert's revision. The court and the opposing party need enough information to assess whether the AI-assisted analysis is reliable and whether it has been adequately verified by a qualified professional.
Can AI-produced analysis be cross-examined?
The AI tool itself cannot be cross-examined: it has no legal standing and cannot give oral evidence. The expert who has used AI and adopted its output is cross-examined on their understanding of the analysis and their ability to verify it. An expert who cannot explain in their own words the methodology applied by an AI tool, or who cannot demonstrate that they verified the AI output against source documents, will face serious difficulty under cross-examination. This is the core practical reason why forensic accountants should not allow AI tools to perform or draft the substantive analysis in their reports: they cannot be interrogated at trial in the way that the expert's own reasoning can be.
What is the CJC's current position on AI in expert reports?
As at 2026, the Civil Justice Council's position, expressed in its 2023 interim guidance and 2024 updated guidance, is that AI use by expert witnesses must be transparent. The CJC does not prohibit AI use but requires disclosure of material AI use and expects experts to verify AI output before relying on it in reports. The CJC is expected to publish more detailed guidance in 2026 as part of its full AI and courts report. The direction of travel is towards more specific rules, and experts and solicitors should assume that requirements will become more prescriptive rather than less as the guidance develops.
Does using AI breach CPR Part 35?
Using AI appropriately, with disclosure and adequate verification, does not breach CPR Part 35. Using AI to draft or formulate the substantive analysis or conclusions of an expert report without disclosure, or without adequate verification, potentially breaches the expert's duty of candour to the court and the requirement that the report represent the expert's own reasoned professional opinion. The breach is not in using AI: it is in certifying under PD35 that the report contains the expert's true opinion when a material part of that opinion was in fact generated by an AI tool that the expert did not adequately verify and did not disclose.
Do all courts in England and Wales require AI disclosure?
As at 2026, there is no single universal rule requiring AI disclosure in all proceedings. However, the CJC guidance applies across civil proceedings in England and Wales, and specific court divisions, including the Business and Property Courts, have issued their own guidance requiring AI disclosure in witness statements and expert reports. The safest professional approach for any forensic accountant is to disclose material AI use in every report regardless of the specific jurisdiction, because the underlying obligation derives from CPR Part 35 and the duty of candour rather than from any specific AI-disclosure rule. Courts that have not issued specific guidance still expect expert witnesses to be candid about the basis of their opinions.
The integration of AI tools into forensic accounting practice is ongoing and will accelerate. The professional and procedural framework in which forensic accountants operate in England and Wales has responded to that development by requiring transparency rather than prohibition. A forensic accountant who understands what AI tools can and cannot do reliably, who uses them for functions where their output is verifiable and limited in scope, and who discloses that use clearly in their expert witness reports, is both professionally compliant and more efficient than one who either refuses all AI assistance or uses it without disclosure. From practice across 150+ instructions, the clearest lesson is that courts respond well to expert transparency: an expert who explains their methodology, including their AI tools, with clarity and specificity is a more credible witness than one who is evasive about any aspect of how their analysis was formed.
To discuss AI disclosure and expert witness standards in forensic accounting instructions for your proceedings, contact Key Ledgers at Key Ledgers enquiry page or call 020 8907 9218.
Author: Bharat Varsani FCCA, CPR Part 35 qualified forensic accountant expert witness, 150+ instructions in England and Wales.