Artificial intelligence can save time in medico-legal reporting. It can also produce errors that are difficult to notice because the prose reads so well.
Its proper role is to support the expert’s work, not replace the clinical reasoning for which the expert remains personally responsible.
What AI can do well.
Large medical bundles contain repetition, inconsistent formatting and prolonged periods with little relevance to the instruction. A secure AI system may help extract dated events, organise records, identify repeated references to symptoms and prepare a draft chronology. It may also assist with formatting, grammar and duplicated wording. Used carefully, these functions can leave the expert more time to consider diagnosis, causation, prognosis and conflicting evidence. The mistake is assuming that speed of production proves accuracy.
AI may fail to distinguish the claimant’s account from a clinician’s finding, overlook an entry that weakens the proposed opinion or give undue prominence to copied text. The expert must therefore return to the original records rather than review only the generated summary.
The opinion must remain the expert’s own.
Practice Direction 35 requires expert evidence to be the independent product of the expert, to remain objective and to take account of material facts that may detract from the opinion. The report must also identify where insufficient information prevents a definite conclusion.
AI cannot assume those duties. It cannot conduct a clinical examination, resolve disputed facts or explain why one diagnosis is preferable to another. The more subtle danger arises when AI drafts the reasoning and the expert edits it. Once a polished explanation appears on screen, it can become easier to approve the wording than to form the opinion independently.
Human review must mean more than correcting spelling and checking names. The expert should verify each material fact, consider omitted evidence and rewrite any passage that does not reflect their own reasoning.
Source-linked drafting is safer.
A useful system should allow every extracted event or generated factual statement to be traced to its source.
If the draft states that the claimant had no previous neck symptoms, the expert should be able to see whether an earlier period was missing. If it reports that physiotherapy produced no improvement, the underlying treatment entry should be available immediately. A chronology should be treated as an index to the evidence, not as the evidence itself. The original document remains the reference point whenever a fact is material to diagnosis, causation or prognosis.
Confidentiality cannot be an afterthought.
Medico-legal bundles contain health information, addresses, employment records and details about family or psychological history. Uploading those documents into an unrestricted public chatbot creates an information-governance risk.
The organisation using the system should know where data is processed, who can access it, how long it is retained and whether it is used to train another model. ICO guidance places accountability, transparency, lawfulness, accuracy, fairness, security and data minimisation at the centre of AI processing involving personal data. The guidance is under review following changes made by the Data (Use and Access) Act, so the current position should be checked when a system is adopted or updated.
A secure platform, written processing terms, access controls and defined retention periods should be established before records are uploaded. Pseudonymisation may reduce risk, but it does not remove all data-protection responsibilities. No improvement in drafting speed justifies careless handling of medical records.
Responsibility stays with the clinician.
The GMC states that medical professionals remain responsible for decisions taken when using AI and must work within the limits of their competence. They are also expected to understand the uncertainties and limitations of the technology.
An expert cannot defend an inaccurate report by saying that the software produced it. The signature belongs to the expert, as does responsibility for the contents.
The user should understand what the system is designed to do, where it commonly fails and how the output should be checked. A clinician who cannot explain how an AI-generated chronology was produced should not rely upon it for a material opinion.
An audit trail should record the documents supplied, the generated draft and the changes made before signature.
Transparency is becoming harder to avoid.
As of July 2026, Practice Direction 35 does not contain a specific requirement for experts to declare AI use. The Civil Justice Council has consulted on a proposal that an expert’s statement of truth should identify and explain substantive AI use, excluding administrative functions such as transcription. It also proposed identifying the AI tool used. The consultation has closed. A June 2026 update said that expert evidence remained an area requiring further consideration and that publication of the final report was anticipated later in 2026. The proposal is therefore not yet a procedural rule.
Even without a specific rule, disclosure is sensible where AI has materially selected, summarised, interpreted or generated content. Routine spelling correction is different from drafting a chronology, identifying inconsistencies or suggesting causation wording.
Transparency allows the expert to explain what the system did and how the output was verified.
Support needs clear boundaries.
AI may organise records, highlight possible gaps and prepare descriptive text. It should not make final decisions about diagnosis, credibility, causation or prognosis.
Those decisions require professional knowledge, direct assessment where appropriate and an understanding of how the evidence fits together. They also require the expert to recognise uncertainty rather than allow software to fill it with confident language.
The best AI-assisted report should still contain the individual expert’s reasoning. Technology may reduce clerical work and improve consistency. It should not standardise professional judgement into automatically generated conclusions. AI has a proper place in medico-legal report writing: beside the expert, handling defined supporting tasks under close supervision.
The moment the system begins to form the opinion, and the clinician merely approves it, support has become replacement.
