The Use of AI in Expert Reports: Implications of the Civil Justice Council Consultation and Emerging Judicial Expectations

Executive Summary

The recent consultation by the Civil Justice Council on the use of artificial intelligence in preparing court documents represents one of the most significant developments for expert witnesses since the formalisation of expert duties under Part 35 of the Civil Procedure Rules. The consultation specifically includes expert reports and explores whether experts should be required to disclose the nature and extent of AI assistance used in their preparation. (Courts and Tribunals Judiciary)

For medical experts, this is not merely a technical issue. It raises fundamental questions regarding:

• Independence of expert opinion.

• Transparency of methodology.

• Data confidentiality.

• Disclosure obligations.

• Professional credibility.

• The future role of human expertise in litigation.

While AI offers remarkable opportunities to improve efficiency, identify inconsistencies within records and assist with document review, the courts are moving towards a position that any use of AI must be transparent, explainable and capable of judicial scrutiny. Failure to disclose material AI involvement may ultimately become as problematic as failing to disclose reliance upon a medical text, guideline, colleague, or external expert.

The Direction of Travel

The CJC consultation considers whether experts should be required to identify and explain any AI used in preparing reports. Several professional bodies responding to the consultation have supported greater transparency regarding AI use. (SLS)

The emerging distinction appears to be:

Generally acceptable

• Spell checking.

• Grammar correction.

• Formatting.

• Transcription.

• Administrative assistance.

Potentially declarable

• Summarising records.

• Chronology generation.

• Literature searching.

• Drafting substantive sections.

• Identifying inconsistencies.

• Producing opinions or conclusions.

Potentially problematic

• Generating expert opinions.

• Assessing breach of duty.

• Assessing causation.

• Producing final conclusions without expert verification.

• Creating evidence not independently verified by the expert.

The proposed approach recognises that AI may assist experts but must not replace expert reasoning. (SLS)

Why Expert Witnesses Are Different

An expert witness occupies a unique legal position.

The court does not instruct an expert because they can write.

The court instructs an expert because they possess specialised knowledge, experience and judgement.

The expert's value lies in:

• Clinical interpretation.

• Weighing competing evidence.

• Understanding uncertainty.

• Applying professional standards.

• Forming independent opinions.

An AI system may review 10,000 pages of records within minutes.

However, the court is not interested in what the AI thinks.

The court is interested in what the expert thinks.

Consequently, any AI involvement that materially contributes to the expert's opinion may become a matter that the court expects to understand.

Could AI Review Records Better Than a Human?

This is perhaps the most interesting question.

In certain respects, the answer may be yes.

AI can:

• Review large volumes of documentation rapidly.

• Detect inconsistencies in dates.

• Identify missing records.

• Highlight contradictory witness accounts.

• Generate timelines.

• Flag unusual patterns.

In a 5,000-page medical negligence case, AI may identify discrepancies that a human reviewer might overlook simply through fatigue.

The problem is not capability.

The problem is reliability.

An AI may identify:

"Record A contradicts Record B."

This can be extremely useful.

However, AI may also:

"Infer" a contradiction that does not exist.

Or worse:

Generate a completely fictional interpretation.

This phenomenon remains known as hallucination.

Recent cases internationally have demonstrated courts imposing sanctions where AI-generated authorities or factual assertions were submitted without verification. The consistent judicial message has been that responsibility remains with the human professional. (Reuters)

The Emerging Standard: AI as a Junior Assistant

The safest conceptual model is to view AI as a highly efficient but inexperienced junior assistant.

You would not:

• Sign a report drafted entirely by a trainee without checking it.

• Adopt another doctor's opinion without review.

• Rely upon a chronology prepared by somebody else without verification.

The same principle should apply to AI.

Every statement included in an expert report must ultimately become the expert's own evidence.

The Disclosure Question

One of the most important practical issues concerns disclosure.

A future expert declaration may resemble:

"Artificial intelligence software was used to assist with chronology generation and summarisation of disclosed records. All outputs were independently reviewed and verified by the expert. No expert opinion or conclusion was generated by AI."

This type of declaration would reassure the court that:

1. AI was used.

2. The scope of use is known.

3. The expert retained responsibility.

The greater the AI contribution, the greater the likelihood of scrutiny.

The Hidden Risk: Prompt Disclosure

A fascinating issue has recently emerged internationally.

Courts have begun considering whether AI prompts may themselves become disclosable as part of an expert's methodology. One recent U.S. dispute concerned whether prompts used by an expert should be disclosed to allow scrutiny of how AI was used in the expert's work. (Reuters)

This creates potentially profound implications.

If prompts become disclosable, parties may seek disclosure of:

• Instructions given to AI.

• Questions posed.

• Uploaded documents.

• Generated outputs.

• Editing history.

Experts may therefore need to preserve AI audit trails in the same way that they preserve notes and working papers.

Confidentiality and GDPR Concerns

Medical experts face particular challenges.

Many publicly available AI systems process information outside the direct control of the user.

Questions arise regarding:

• Patient confidentiality.

• GDPR compliance.

• Data processing agreements.

• International data transfer.

An expert who uploads complete medical records into an unsecured public AI platform may face criticism regardless of whether the resulting report is accurate.

Future best practice is likely to favour:

• Enterprise AI systems.

• Secure NHS-compliant platforms.

• On-premises AI solutions.

• Formal information governance policies.

Confidentiality breaches may ultimately prove a greater risk than hallucinations.

What Happens if an Expert Does Not Disclose AI Use?

This may become the most dangerous area.

Suppose an expert:

• Uses AI extensively.

• Generates chronologies.

• Produces summaries.

• Drafts significant sections.

Yet declares nothing.

If discovered during litigation, opposing counsel may argue:

Credibility

The expert has failed to be transparent.

Independence

The opinion was not wholly the expert's own.

Methodology

The methodology cannot be scrutinised.

Disclosure

Relevant material has been withheld.

Professional Conduct

The expert failed to comply with evolving professional standards.

Even if the opinion itself is correct, the expert's credibility could suffer significant damage.

The greatest danger may not be that the report is wrong.

The greatest danger may be that the expert appears evasive.

Courts tend to forgive honest disclosure more readily than undisclosed methodology.

A Practical Framework for Medical Experts

I believe the safest future approach is:

Level 1: Administrative AI

Examples:

• Spelling.

• Grammar.

• Formatting.

Disclosure probably unnecessary.

Level 2: Organisational AI

Examples:

• Chronologies.

• Record summaries.

• Timeline generation.

Maintain records of use and consider disclosure.

Level 3: Analytical AI

Examples:

• Identification of inconsistencies.

• Literature searching.

• Evidence mapping.

Disclosure advisable.

Independent verification essential.

Level 4: Opinion-Generating AI

Examples:

• Breach of duty analysis.

• Causation opinions.

• Prognosis opinions.

Avoid or disclose comprehensively with extensive verification.

Likely Future Position

The future expert declaration may eventually include an AI section analogous to declarations of truth and statements of independence.

Experts may be required to specify:

• Whether AI was used.

• Which system was used.

• What tasks it performed.

• Whether confidential information was uploaded.

• What verification steps were undertaken.

This would mirror the transparency expected in other aspects of expert methodology. (Law Society)

Conclusion

AI is unlikely to be prohibited from expert witness practice.

Indeed, it may become indispensable.

A well-designed AI system may eventually review records more thoroughly than any individual expert and identify discrepancies that would otherwise remain hidden.

However, the legal system is making it increasingly clear that the expert cannot delegate responsibility.

The expert may use AI.

The expert may benefit from AI.

The expert may even rely upon AI-assisted analysis.

But the opinion must remain the expert's own, and the methodology must be capable of explanation, scrutiny and disclosure.

The emerging judicial principle can be summarised in a single sentence:

AI may assist the expert, but it cannot become the expert.

For medical experts preparing for the next decade of litigation, transparency, auditability and independent verification are likely to become just as important as the technology itself. (Courts and Tribunals Judiciary)