The expert explains that an artificial intelligence system reviewed the records, selected the entries and drafted several sections. The expert checked the finished document but retained no record of the system’s output and did not mention its involvement in the report.
The problem is not necessarily that AI was used. It is that nobody was told what work it performed.
The rules are moving towards disclosure.
As of July 2026, Practice Direction 35 does not contain a specific declaration requiring an expert to identify the use of AI. It does, however, require the report to set out the material relied upon, identify the facts and instructions supporting the opinions, and distinguish matters within the expert’s own knowledge. The expert must confirm that the opinions expressed represent their true and complete professional opinions.
The absence of a specific AI rule should not be mistaken for approval of undisclosed use.
In February 2026, the Civil Justice Council proposed amending the expert’s statement of truth so that an expert would explain any use of AI other than transcription or similar administrative assistance and identify the tool used. The proposal expressly recognised that expert evidence may properly include AI-generated material, provided the involvement of the system is identified and explained.
That consultation has closed, but the final report has not yet been published. In its June 2026 update, the Civil Justice Council described its emerging approach as one of proportionate transparency for expert evidence. Publication of the final recommendations is anticipated later in 2026. The proposed disclosure requirement is therefore not yet an amendment to the Civil Procedure Rules, but the direction of travel is plain.
Not every automated function needs a declaration.
A useful dividing line is whether the technology generated or influenced substantive content. An expert should not need to declare that software corrected spelling, adjusted formatting or transcribed dictated words without changing them. Such functions are comparable to ordinary administrative support.
The Civil Justice Council’s interim proposal draws a similar distinction between administrative functions and systems that generate substantive material.
The position changes once AI begins to influence the substance of the report rather than merely its presentation. This includes extracting events from medical records, selecting which entries appear relevant, summarising the claimant’s previous history and highlighting apparent inconsistencies. The same concern arises where the system drafts examination findings or opinion sections, suggests a diagnosis, prognosis or causation analysis, or rewrites the expert’s reasoning in more persuasive language. At that point, AI is no longer performing a purely administrative function; it is contributing to the evidence placed before the court.
The number of words generated is not the proper test. A system may produce a lengthy but source-linked chronology that the expert checks carefully. Alternatively, it may contribute a single sentence containing the central causation opinion, here what matters is whether the AI affected the evidence placed before the court.
Signing the report does not answer the authorship question.
Some experts will say that disclosure is unnecessary because they remain responsible for the finished report. The expert’s signature confirms responsibility. It does not explain how the report was prepared or whether the reasoning was independently formed.
An AI system may omit an adverse entry, merge two consultations or present an allegation as an established clinical fact. It may generate a polished explanation that was not apparent from the records. If the expert merely reads that explanation and agrees with it, the court may ask whether the opinion was formed by the expert or adopted from the machine.
The GMC expects medical experts to give objective and impartial evidence within their competence. Medical professionals must take reasonable steps to check the accuracy of their evidence and be able to explain and justify their decisions. Its guidance on AI also makes clear that clinicians remain responsible for decisions made when using modern technologies and must understand their uncertainties and limitations. Human review must therefore involve more than proofreading. The expert should return to the source records, check every material statement and form the diagnosis, causation opinion and prognosis personally.
“I read it and agreed with it” is a poor explanation for an expert opinion.
What a proportionate declaration should say.
Disclosure need not become a technical appendix describing the architecture of the software. The court usually needs to know what the system did and how the expert controlled the risk.
A declaration might state:
AI-assisted software was used to organise the medical records and prepare a draft chronology. I checked each material entry against the original records and amended or removed any inaccurate or incomplete content. The clinical assessment, diagnosis, causation opinion and prognosis are my own.
Where the system drafted substantive passages, that should be said expressly. Where it suggested diagnoses or identified inconsistencies, the expert should explain that those suggestions were independently assessed.
A vague statement that “technology assisted in preparing the report” may conceal more than it reveals.
Nor should an expert describe substantive drafting as transcription. Transcription reproduces the expert’s words. Generative AI creates new wording and may introduce new reasoning. The distinction will matter if the report is challenged.
Disclosure does not make unsafe processing lawful.
An expert may disclose the use of AI and still have handled the claimant’s information improperly.
Medico-legal bundles contain detailed health information and other sensitive personal data. GMC confidentiality guidance requires medical professionals to use the minimum information necessary and protect personal information against improper access, disclosure or loss. Those responsible for information systems must understand who can access the information and how it is processed. The ICO’s AI guidance requires organisations using personal data in AI systems to consider lawfulness, transparency, accuracy, fairness, security and data minimisation.
Identifiable records should not be pasted into an unrestricted public chatbot without an approved data-processing arrangement. Before using any system, the expert or commissioning organisation should establish where the information is stored, whether it is retained, who can access it and whether it is used to train a wider model.
Openness is the safer professional position.
There may be concern that disclosing AI use will invite unnecessary questions. That is possible. Concealing material use creates more difficult questions. If the system was secure, the sources were checked and the opinion was independently formed, the expert should be able to explain that calmly. If the expert cannot explain what the system did, how its output was verified or which passages it drafted, the problem is not the declaration. It is the method used to prepare the evidence.
Experts should disclose AI when it materially selects, summarises, interprets or generates content. They do not ordinarily need to declare routine spelling, formatting or faithful transcription. The expert’s duty remains personal. Software cannot sign the statement of truth, attend the experts’ discussion or answer questions in the witness box. AI may assist in writing the report however, the court is still entitled to know who, or what, helped produce the evidence it has been asked to trust.
