The courts are beginning to put a name to a duty they have always implied: that where AI touches legal evidence, its use should be transparent to a proportionate degree. The Court of Appeal in R v FGD [2026] EWCA Crim 918, the Civil Justice Council's current review and the BSB's guidance on AI all converge on the same idea. For an expert witness, proportionate transparency means being able to say what a tool did, what the expert checked, and what stayed the expert's own judgement. This post explains what the courts are asking and how to meet it in practice.

The practical bottom line

  • Use AI only for the reading and organising of records that can be checked against the source page.
  • Keep a short record of the tool, what it was used for, and how the output was verified.
  • Keep the clinical judgement and the opinion as the expert's own, under the normal Part 35 duty.
  • Assume transparency will be expected even before the CJC settles a formal rule, and prepare for it now.

What does proportionate transparency mean?

Proportionate transparency is the idea that AI use in legal evidence should be disclosed to a degree that matches the significance of the evidence and the risk of it being misused. A summary an expert used only to orient themselves in a 5,000 page bundle needs less explaining than a tool that helped shape the reasoning of a report. The point is not to ban AI, and it is not to force a formulaic confession in every case. It is to make the use of AI visible enough that the court, the instructing solicitor and the other side can tell what the expert actually did.

For an expert the practical shape of that duty is a short, honest account: which tool was used, what it was used for, what the expert verified against the records, and what judgement remained their own. Written down at the time, that account answers almost every question before it is asked, and it is far stronger than a reconstruction attempted months later.

Why are the courts focused on this now?

The trigger is a string of decisions and guidance in 2026. In R v FGD the Court of Appeal held that using AI to coach a witness could amount to prohibited practice, and the reasoning has been read as carrying into civil litigation. The Civil Justice Council is exploring proportionate transparency for witnesses' use of AI, and the BSB has clarified how its existing handbook duties apply to AI at the Bar. None of these creates a bespoke AI rule for experts. Together they signal that the courts will treat the line between assist and shape as the thing to watch.

The concern is not that an expert used a tool to read the bundle faster. It is that AI might quietly be doing work the expert is supposed to do, or quietly be supplied with records in a way no one can account for. Transparency is the courts' way of keeping the expert's judgement where the law, and the statement of truth, put it.

Does this change an expert's duties under Part 35?

No. The duties in Part 35 already cover this. The expert's overriding duty is to the court, the report must be the expert's independent product stating the substance of the instructions and the basis of each opinion, and the expert signs a statement of truth. Practice Direction 35 requires the facts, assumptions and material relied on to be stated. None of this needed amending to reach AI, because it is written around responsibility rather than method. Proportionate transparency is not a new rule; it is the courts making explicit what those duties were already asking.

What should an expert actually record?

A short, contemporaneous record covers the ground:

  • Which tool and version was used.
  • What it was used for, such as extracting typed facts, building a first chronology, or locating a set of entries.
  • What it produced, and what the expert verified against the source page before it informed the report.
  • That the clinical judgement and opinion remained the expert's own.
Transparency is not proved by a blanket confession. It is proved by a trail of what was used, what was checked, and what stayed the expert's own.

If the tool attaches a source page to every fact it returns, the verification record largely writes itself, because every fact that entered the report carries the page the expert checked it against. Where the tool runs on controlled infrastructure that logs every step, the audit trail exists independently of the expert having to remember it.

What is the line between assist and shape?

This is the distinction the courts keep returning to, and it is the most useful test for an expert. A tool that reads the bundle, extracts facts and builds a first chronology is assisting: its output is checkable against the records and the expert decides what it means. A tool that drafts the reasoning, weighs the evidence or forms a view on breach or causation is shaping the evidence, and that is where the judgement belongs, by definition, to the expert alone.

The safe position is simple: use AI for the reading and organising that can be verified, and keep the clinical judgement as your own. That division of labour is covered in more depth in Can an expert witness use AI? Part 35, responsibility and disclosure, and the tooling side on the AI medical record review page.

How should an expert manage patient data with AI?

Transparency cuts both ways, and so does the law. Medical records are special category data, and the ICO's guidance on AI and data protection applies in full. An expert whose records go through a tool has to know where they sit, who can see them, and whether every access is logged. A tool that quietly posts a patient bundle to an external service is itself the opposite of transparency, however fast its answers are. The governance of the data and the transparency of the method are two halves of the same duty.

Why it matters

The courts are not asking experts to stop using AI. They are asking experts to be able to say what they did with it, in a way that fits the case, and to keep the judgement where it belongs. Meeting that expectation is straightforward when the record-keeping is built into the workflow rather than bolted on afterwards. That is the principle ALLDOQ is built on. It reads the whole record, extracts the facts, builds a chronology and answers questions with the source page attached, on infrastructure the organisation controls, with every step logged for audit. The expert gets the bundle read in hours and signs a report in which every fact can be traced to the page it came from, so proportionate transparency is not a question the expert has to scramble to answer. It is already part of how the work was done.

Sources