In January 2026, Mr Justice Waksman used a phrase that should have got every expert witness’s attention. A solicitor had handed an expert a pre-drafted AI generated report before the expert had done their own analysis. The judge called it a gross breach of duty. He didn’t stop there. An expert who takes AI generated opinions and puts their own name to them, he made clear, is just as exposed as the solicitor who handed the report over in the first place.
That warning landed a month before the Civil Justice Council opened a consultation asking a question that now looks inevitable in hindsight: should expert witnesses have to declare, in writing, exactly how AI touched their reports.
The proposal on the table is specific. Amend Practice Direction 35, the rule that governs expert evidence, so that a report has to state which AI tools were used and for what purpose. Not a ban. Not a blanket declaration that no AI was involved anywhere in the process. A disclosure, sitting inside the statement of truth, of precisely where the tool did and didn’t touch the analysis.
The consultation closed in April 2026. In July, the Council published an update on where the responses had left things, and the shape of the final rule started to come into focus. For most legal drafting, skeleton arguments, statements of case, the kind of documents lawyers produce every day, the Council decided existing professional rules were doing the job fine. No new AI specific requirements needed there. Expert evidence was treated differently. It was singled out, on its own, for what the Council called proportionate transparency. A final report is expected later this year, but the direction of travel is no longer really in doubt.
Why single out experts. Because of what an expert report actually is. A solicitor’s skeleton argument is legal argument, tested and challenged by the other side as a matter of course. An expert report is different. The court is relying on it as an independent professional opinion, not a piece of advocacy, and the whole system of expert evidence under CPR Part 35 rests on the idea that the opinion in front of the judge is genuinely the expert’s own. That is the exact assumption Waksman’s warning was protecting, months before the Council put a formal proposal behind it.
The line the courts are drawing isn’t about whether AI touched a report at all. It’s about where. Using a tool to summarise literature, tidy formatting, or transcribe an interview sits on one side of that line. Using it to help form or express the opinion the expert was instructed to give sits on the other, and that side is where Waksman’s gross breach of duty language starts to apply. What the CJC’s rule would add is a mechanism forcing that distinction into the open, in writing, every time.
For forensic psychiatrists instructed as experts, the sensible move is to stop treating this as a future problem. Keep a record now of any AI tool used at any stage of a report, what it did, and when. Be clear, for your own sake as much as anyone’s, about which side of that line each use falls on. And raise it with instructing solicitors at the letter of instruction stage, so expectations are set before the report is written rather than defended after the fact.
It’s the same authorship question we looked at with Godwin v Godwin last time, just formalised into a rule and pointed at the professionals whose opinions carry the most weight in the room.
If you’d like help putting a clear, defensible AI use policy in place before this becomes a formal requirement, get in touch with our team. We can talk through what good practice already looks like for expert reports.