Do Expert Witnesses Need to Disclose AI Use in Reports? The 2026 Regulatory Picture

Nobody is banning AI from expert witness work. But the question of AI use in expert witness reports has moved a long way in the last eighteen months, from a background ethical debate to an actual consultation with a Practice Direction attached, and if you're an expert witness, or you instruct one, that shift matters more than it might first appear.

What the Civil Justice Council actually proposed

In February 2026 the Civil Justice Council published an interim report and consultation on how AI should be handled in court documents. The central proposal was straightforward on paper: any expert who uses AI in preparing a report, beyond pure transcription or basic admin, should have to disclose which tool they used and what for. The mechanism for that would be an amendment to Practice Direction 35, the rule that already governs what an expert report must contain and how an expert must certify it.

This didn't come from nowhere. Judgments in 2025, including one from Dame Victoria Sharp, President of the King's Bench Division, had already criticised cases where experts or their legal teams had relied on AI-generated citations that turned out not to exist. Fake authorities appearing in a witness statement are exactly the kind of thing that makes a regulator move quickly, and the CJC did.

Where AI use in expert witness reports stands now

The consultation closed on 14 April 2026, and by July the CJC had published an update on the responses it received. Lawyers drafting pleadings or skeleton arguments, it concluded, don't need new AI-specific rules, since existing conduct rules already cover it. Expert evidence is being treated differently. The direction of travel there is toward a proportionate transparency requirement: an expert will need to explain whether AI played any part in preparing their report, and if it did, name the tool and describe what it was used for. Witness statements are still being debated separately, and a final report is expected later in 2026, so nothing is fully settled yet, but the pattern is clear enough that waiting for the final wording before thinking about it would be a mistake.

Why experts are already cautious

The regulatory pressure lands on ground that was already cautious, which is exactly what makes AI use in expert witness reports such a sensitive subject right now rather than a purely theoretical one. Bond Solon's 2025 Expert Witness Survey, 525 UK experts fielded in September and published that November, found only 20% had used AI in their expert witness work at all. That's double the year before, but still a small minority, and well below the 65% general UK workforce AI-usage figure reported elsewhere. More tellingly, 86% said they would refuse an instruction if a solicitor supplied them with an AI-drafted report, and 89% want clearer guidance before they'd trust AI use any further.

That's not a profession waiting to be told AI is fine. It's one actively drawing its own line ahead of the regulator drawing one for them, and the two lines look likely to end up in roughly the same place.

What disclosure will likely mean in practice

If the proportionate transparency approach the CJC currently favours makes it into the final Practice Direction 35 amendment, an expert report will need a line, or a short section, stating whether AI was used anywhere in its preparation and, if it was, naming the tool and describing what it did. Recording that a voice-to-text tool captured an interview is a very different disclosure from admitting a drafting tool wrote part of the opinion itself, and reports will need to be specific enough to make that distinction clear rather than folding everything under one vague AI mention.

That specificity is where accurate documentation earns its keep. An expert who has used a proper transcription service for interviews and consultations, with a human typist doing the actual conversion from speech to text rather than an AI tool, has a straightforward answer ready when the disclosure question comes: no AI was used at the capture stage, because a person did it. That isn't a workaround for the new rules. It's simply what the disclosure requirement is designed to surface, stated plainly because it happens to be true.

For experts weighing up where AI genuinely helps and where it creates a disclosure headache later, the safest ground is close to where Bond Solon's own respondents already draw their line. AI for the administrative undergrowth, tidying language and chasing references, is one thing. The opinion and the analysis behind it need to stay entirely the expert's own work, checked and produced by a person from the recording onward. OutSec Medical's transcription service sits on the human side of that line, which is a straightforward thing to be able to say when a report gets scrutinised and someone finally asks the question directly.

If you'd like to see how that works in practice, from recording to finished transcript, our medico-legal transcription services page sets out exactly what's involved. Get in touch if you'd rather talk it through first.