Artificial intelligence can produce a convincing clinical summary, a persuasive explanation ofcomplex evidence, or a professionally drafted report within seconds. But what happens when theinformation is wrong? What if a clinical finding is omitted, a reference is entirely fabricated, or anAI-generated explanation subtly warps your interpretation of the evidence?

The consequences are not simple technical errors. Within medico-legal practice, they directlythreaten an expert’s credibility, the reliability of evidence, the progress of litigation, and the lives ofthe individuals before the court. The question is no longer whether AI can assist expert witnesses,but whether professionals understand the risks well enough to use it without compromising theirindependence and duty to the court.

When Convincing Information Cannot Be Trusted

AI does not have to produce obviously nonsensical information to create a problem, its mostdangerous failures appear entirely credible. A generated chronology may omit a vital entry inmedical records. A summary may present an allegation as established fact. A research tool may citea non-existent study or misrepresent what a genuine publication concluded. Even using AI purely tosharpen the wording of a report can inadvertently alter the substance of your original opinion.

These are practical, active risks. In R (Ayinde) v Haringey LBC [2025] EWHC 1383 (Admin), the HighCourt addressed fabricated legal authorities generated by AI tools. The Solicitors RegulationAuthority subsequently highlighted this alongside other definitive cases, including BCP v A Mother[2026] EWFC 71 (B) and Cork v Smith [2026] EWHC 1199 (Ch), in its August 2026 warning notice.

While these specific cases concerned legal submissions rather than findings against expert witnesses,they demonstrate exactly how severely the courts view unchecked AI "hallucinations" entering legalproceedings.

Parallel to this, the Civil Justice Council’s consultation on AI in court documents has identifiedexpert evidence as an area requiring urgent consideration and stricter safeguards. If an AIgenerated error compromises your report, can its origin be identified, its accuracy proved, and yourresulting opinion independently justified under cross-examination?

The Risks Extend Beyond Accuracy

Accuracy is only one facet of the danger. Confidential medical records or sensitive litigationmaterial may be exposed through the inappropriate use of public AI platforms. Notably, the UpperTribunal ruled in UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC) that pastingclient information into open-source AI tools like ChatGPT effectively places that data on theinternet in the public domain.

Furthermore, automation bias frequently causes professionals to place undue confidence in a neatlygenerated answer. Critical contextual details are stripped away during summarisation, while animpeccable narrative can make an unsupported conclusion appear far more authoritative than theunderlying evidence warrants. Most insidiously, an expert may believe they are exercising independent judgment, unaware that their perspective has already been skewed by a biased AIsummary or a selective presentation of facts.

The judiciary is monitoring this closely. The Position Statement on the Judicial Use of Artificial Intelligence,published on 9 October 2026, reinforces that AI must never displace human reasoning, evidenceevaluation or accountability. Under CPR Part 35 and Practice Direction 35, expert evidence mustremain independent, consider all material facts and disclose any limitations affecting the opinion.

APIL’s response to the Civil Justice Council’s April 2026 consultation specifically supports expertsexplaining their use of AI where it contributes to specialist content in court documents,distinguishing substantive generation from administrative tasks such as transcription. The CivilJustice Council’s subsequent update confirms that expert evidence remains an area requiring furtherconsideration, with proportionate transparency under discussion.

Professional Responsibility Requires More Than Caution

AI should neither be blindly rejected nor adopted simply for speed. A defensible approach mustbegin before the platform is opened. The expert must establish the precise purpose of the task,evaluate whether AI is appropriate, and determine whether its contribution risks altering theirindependent reasoning.

Checking the output requires more than a casual final read-through of text that already looksconvincing. Experts must rigorously cross-reference every fact, reference, and clinical finding againstreliable primary sources. They must ensure that summaries have not distorted the evidence, andthey must remain entirely capable of explaining their exact methodology if challenged.To maintain a defensible standard of practice, you must remain in control throughout, recognisingwhen AI can safely assist, when exhaustive human verification is required, and when a task demandsindependent professional judgment alone

Ultimately, AI cannot take responsibility. The professional does.

My book, AI in Professional Practice: A Defensible Framework, explains and explores these issues in greaterdepth and in plain language. While proposed government regulations and frameworks are notexpected for months, this book provides the practical governance protocols and codes of practicethat can be implemented immediately. It shows you how to use the robust framework I designed tohelp professionals confidently navigate their AI tools now, while it guides you on how to mitigate riskwhile firmly retaining your professional judgment and accountability to clients.

www.theaiframework.co.uk

Sian Francis

‍