AI in expert evidence: Ireland shows where the courts are heading

The debate around artificial intelligence in litigation is beginning to move beyond the question of whether lawyers and experts should use AI. A more important question is emerging: if AI is used, can its output be independently verified?

A new Practice Direction issued by the President of the Irish High Court provides one of the clearest indications yet of how courts may approach that question. Coming into force on 1 September 2026, the Practice Direction on the Responsible Use of Generative Artificial Intelligence in Court Documents applies to pleadings, submissions, affidavits, witness statements and, importantly for the medico-legal sector, expert reports.

And it is not an anti-AI document. In fact, the High Court expressly recognises that generative AI may help make litigation more efficient, reduce costs and improve access to justice. The emphasis instead is on verification, transparency and human responsibility.

AI does not transfer responsibility

Perhaps the most important principle in the Practice Direction is also the simplest: using AI does not change who is responsible for the finished document. The lawyer, witness or expert remains responsible for what is submitted to the court.

Where generative AI has contributed to a court document, its output must therefore be independently verified under human supervision. Simply asking another AI system whether the first AI system was correct is specifically identified as insufficient.

That distinction matters. The problem with AI in litigation is not simply that an algorithm has been involved. It is whether a professional can establish where the information came from, whether it is accurate and whether it genuinely supports the conclusion being expressed.

The reference to chronologies is particularly significant

Buried within the requirements relating to pleadings is a short but important provision: chronologies must be checked for accuracy.

For clinical negligence and personal injury litigation, this is highly relevant. Medical-record chronologies can involve thousands of pages of source material and hundreds or thousands of individual clinical events. AI offers an obvious opportunity to reduce the considerable time and cost involved in producing them. But a chronology is only useful if its individual entries can be checked against the underlying evidence.

That means provenance matters. Source documents matter. Page references matter. And the ability to move from an extracted event back to the underlying medical record matters. The future is unlikely to belong to systems which simply produce persuasive-looking prose and ask users to trust it.

Expert reports receive even greater scrutiny

The Practice Direction goes further when dealing with expert evidence. An expert report must remain the product of the expert’s own opinion, knowledge and reasoning. Where generative AI has been used, the expert must disclose the nature and extent of that use so that the court and the parties can assess the reliability, independence and evidential weight of the opinion.

The Direction specifically identifies material AI use in drafting, analysing, interpreting or generating substantive content as requiring disclosure. This is an important distinction for medico-legal practice. There is a fundamental difference between using technology to:

• organise thousands of pages of medical records;
• identify dates, investigations and clinical events;
• locate relevant passages in the source material; and
• assist an expert in navigating a large evidential bundle;

and asking a general-purpose AI system to formulate the expert’s medical opinion. The first can augment professional analysis. The second risks obscuring who is actually providing the expert evidence. Good medico-legal AI should make that boundary clearer, not blur it.

England and Wales is considering the same questions

This development should not be dismissed as relevant only to Irish proceedings. The Civil Justice Council in England and Wales is already examining the use of AI in preparing court documents, including expert reports.

Its June 2026 update is particularly interesting. While there appears to be broad support for relying on existing professional duties when lawyers use AI for conventional legal drafting, the CJC identified expert evidence as an area requiring further consideration. Its emerging direction includes the possibility of “proportionate transparency” in relation to expert evidence. A final report is expected later in 2026.

The direction of travel is therefore becoming clearer. Courts appear willing to accept that AI will become part of professional practice. But acceptance of AI is being coupled with expectations around human oversight, verification, transparency and accountability.

The real dividing line may be verifiability

Much of the discussion about AI in law has focused on hallucination. That is understandable. Courts in multiple jurisdictions have now encountered fabricated authorities and other inaccurate material generated by large language models. But hallucination is only part of the issue. For medico-legal work, the deeper question is whether the technology has been designed around the evidential nature of the task.

• If an AI system identifies an important consultation from a 5,000-page medical bundle, can the expert immediately inspect the original record?
• If it creates a chronology entry, can that entry be traced back to its source?
• If it summarises an investigation or diagnosis, can the professional easily check whether the summary is complete and accurate?
• And does the system assist the expert in reaching their own opinion, rather than attempting to substitute for it?

These are increasingly likely to become central questions when professionals select AI systems for medico-legal work.

AI is not being excluded from the courtroom

The Irish High Court’s Practice Direction is significant precisely because it does not say that AI has no place in litigation. It acknowledges its potential to reduce costs and improve access to justice. But it couples that opportunity with a straightforward principle: AI can assist the professional. It cannot replace the professional’s responsibility for the evidence.

For those developing and adopting AI in medico-legal practice, that is not a barrier to innovation. It is increasingly becoming the specification against which responsible innovation will be judged. And as courts begin turning broad ethical principles into concrete procedural requirements, traceability, verification and human control may prove far more important than simply having the most powerful AI model.