Can medical experts use AI on medical records?
What UK courts, regulators and professional bodies now expect, and what a defensible answer looks like for experts and the agencies that instruct them. Last updated June 2026.
One in five experts already uses AI
The Bond Solon Expert Witness Survey 2025 (525 experts, run with the Law Society Gazette) found that the proportion of expert witnesses using AI doubled in a year, to around 20%, primarily for research, drafting and data analysis. The same survey found that 89% of experts want specific guidance on how AI may be used.
In practice, much of that use happens on consumer tools: discharge summaries pasted into ChatGPT, clinic letters summarised by Claude, GP records run through Gemini, often on personal accounts. These tools were never designed for patient data. The confidentiality, data protection and professional duty exposures sit with the expert, and with the agency that instructed them.
What has changed: the regulatory timeline
- September 2025Bond Solon survey: expert AI use doubles to 20%; 89% call for specific guidance.
- January 2026Mr Justice Waksman, head of the Technology and Construction Court, warns the Bond Solon conference that a solicitor providing an expert with a pre-drafted AI-generated report commits a "gross breach of duty".
- January 2026The Academy of Experts publishes guidance for expert witnesses on the use of AI: administrative use (summarising disclosed documents, preparing chronologies, organising bundles) is accepted; AI must not generate the expert's opinion.
- February 2026The Civil Justice Council's interim report proposes amending Practice Direction 35 to require experts to identify what form of AI has been used in their report.
The direction of travel is clear: AI-assisted preparation is legitimate, and disclosure of it is becoming mandatory. When an expert is asked (by a solicitor, under Part 35 questioning, or in court) what tool touched the records and what it did, "a personal ChatGPT account" is not an answer that survives scrutiny.
Why the tool choice matters
Three categories of tool are in use today, and they fail or pass the defensibility test very differently.
| Consumer AI ChatGPT, Gemini, Claude (personal accounts) |
Generic legal chronology tools US-litigation-focused platforms |
Health Narrator | |
|---|---|---|---|
| Designed for patient data | No | Partlybuilt for law firms, not clinical use | Yesbuilt with clinicians, used across NHS sites |
| UK GDPR / DPA compliance | User's responsibility | Variesoften US-hosted | Yesthird-party audited |
| NHS DSPT registration | No | No | Yes |
| DCB0129 clinical safety certification | No | No | Yes |
| Source-linked citations | Noanswers can't be traced to a page | Partial | Every findingone-click verification at the source line |
| Patient data used for model training | Riskdepends on account tier and settings | Varies | Never |
| Audit trail | No | Varies | Full |
| Adapts to the expert's specialty | Generic | Generic | Yespsychiatry to orthopaedics, same platform |
| Drafts opinions | Will if askedthe core danger | Varies | Never, by designfactual sections only; opinion stays with the expert |
| Cross-references expert reports on a case | No | No | Yesfor joint statements and expert discussions |
What a defensible answer looks like
For an agency, the test is simple to state: if an instructing solicitor or a regulator asks how patient data was handled and what role AI played, can every expert on your panel give the same, documented answer? That requires four things. The tool must be lawful to use on patient data (UK GDPR/DPA, with NHS-grade registration where records originate in the NHS). Every AI-assisted finding must be verifiable at the source: page and line. The platform must keep an audit trail of what was processed and when. And the boundary must be structural, not behavioural: a tool that cannot draft opinion is a stronger answer than a policy asking experts not to.
This is the gap between "our experts don't misuse AI" and "our panel works on a platform where misuse is not possible, and here is the documentation." The first is an assurance. The second is a defence.
Where Health Narrator fits
Health Narrator was built for exactly this boundary: it produces structured, source-referenced chronologies shaped to each expert's specialty, answers plain-language questions about the records with exact page references, drafts factual sections only, and never drafts opinion. It carries NHS DSPT registration, DCB0129 clinical safety certification and audited UK GDPR compliance, with MHRA Class I registration in progress alongside Brighton & Sussex Medical School and The Hill at Oxford University Hospitals. It is already used hundreds of times every week by experts and clinicians working through Expert in Mind, the Royal Free Hospital, The Priory Group and other NHS sites, with users reporting time savings above 60% on record review.