A recent family court judgment has put expert witnesses’ use of AI back under the spotlight. In Cornwall Council v M & Ors [2026] EWFC 306, HHJ Flexman, sitting in the Family Court at Truro, declined to place significant weight on an expert’s evidence after finding that AI had moved beyond presentational help and into generating the substance of a report.
We welcome the judgment. It draws a line that we have built Obiter Review around from the start: AI can safely help an expert check their work, but it must never do the expert’s thinking or writing for them. This article explains where that line sits and why Obiter Review falls on the right side of it.
What went wrong in Cornwall Council v M
The expert, an independent social worker in care proceedings, had used Microsoft Copilot to organise information, summarise interview transcripts and help with drafting. Her first report did not mention AI, and she explained her use of it only once questions were raised. She said Copilot produced better English than she could, and maintained that the conclusions were her own. In oral evidence she confirmed that Copilot summarised her interview transcripts and generated her reports, and that one kinship assessment was entirely AI-generated from her notes [33].
The court’s concerns were specific:
- AI had produced the analysis, the structure, the subheadings and the overall presentation of the report.
- Having heard her give evidence, the judge found the tone and language of the reports very different from her own [35].
- Copilot’s own drafting commentary had been left in, for example noting that a version was “about half the length” and “more likely to withstand cross-examination”. This pointed to inadequate human oversight.
- There were factual errors: the report wrongly said the father accepted the court’s findings, and described a child as autistic without a diagnosis. The report also adopted a specialist risk-assessment model she had not been instructed to use, which the judge suspected was an AI-driven inclusion. The judge could not tell whether these problems came from the notes or from the AI [36].
The judge also had wider concerns about the evidence, which she described as “lacking in rigour and clarity of thought”. Notably, security was not the problem: the judge accepted that Copilot was a closed, secure tool with no risk of publication. The problem was authorship.
The court’s sliding scale
HHJ Flexman accepted that there is a sliding scale from acceptable to unacceptable AI use, and found that generating an entire report from notes strayed into the latter. She accepted that the primary material was the expert’s own, but held that having AI turn it into the report was still a higher-risk approach that affected the weight of the evidence [37]. She drew on the Academy of Experts’ January 2026 guidance, which separates lower-risk from higher-risk uses.
| Lower risk – generally acceptable | Higher risk – affects the weight of the evidence |
|---|---|
| Transcription | Generating substantive content, such as the opinion |
| Spelling and grammar | Material analysis of the evidence |
| Summarising information | Generating an entire report from notes |
The Academy of Experts guidance, set out at length in the judgment, lists quality control among the recognised uses of AI by experts: “using AI to undertake proof-reading or to identify possible errors or inaccuracies in the expert’s work product” [24]. That is the core of what Obiter Review does.
Under Practice Direction 25B, an expert must say who carried out any test, examination or interview used for the report. The judge considered that AI use could be akin to that kind of third-party involvement, depending on how it is used. The question for any expert, then, is not whether a tool uses AI, but what the tool actually contributes to the report.
Why Obiter Review sits at the low-risk end
Obiter Review reads a report the expert has already written and returns a separate, advisory review. It never amends or rewrites a report, never suggests wording and never offers an alternative opinion. Every decision about what, if anything, to change stays with the author.
Set against the concerns in Cornwall Council v M, the difference is structural, not a matter of degree:
| What the court criticised | How Obiter Review works |
|---|---|
| AI generated the analysis and opinions | Obiter Review forms no opinions and offers none. Analysis of the evidence remains the expert’s own. |
| AI generated the structure, subheadings and presentation | The report is complete and structured by the expert before it reaches us. We comment; we do not restructure. |
| The tone and style were not the expert’s | Nothing we produce is inserted into the report, so the voice throughout is the expert’s. |
| AI drafting notes were left in the report | No AI-generated text enters the report, so there is nothing to leave behind. The review is a separate document. |
| Factual errors of uncertain origin | Because AI never writes into the report, it cannot introduce errors into it. The review is designed to find them, by checking the report against its instructions, records and history. |
| Evidence lacking rigour and clarity | The review asks whether conclusions follow from the reasoning given and whether fact is kept separate from opinion. It prompts the expert to look again; it does not supply the answer. |
The judgment contains a telling contrast. Copilot’s leftover notes show it rewriting the expert’s material so that it was more likely to withstand cross-examination. Obiter Review also looks for likely lines of cross-examination, but it points them out and leaves the expert to decide whether, and how, to deal with them. That is the difference between review and authorship.
Put simply, Obiter Review is closer to a careful colleague reading a draft than to a ghost-writer. Independent peer review of a draft report is a long-standing practice, and nobody suggests it compromises authorship provided the opinion remains the expert’s own. Obiter Review makes that kind of scrutiny faster and more structured, without crossing into authorship.
Confidentiality: a separate safeguard
Authorship was the issue in this case, but confidentiality is the other question every expert should ask of an AI tool. Expert reports contain medical records, witness evidence and highly sensitive personal information.
- Reports are processed offline, in England, on hardware owned and used only for Obiter Review.
- No public, consumer or general-purpose AI platform is used, and nothing is sent through one.
- Processing is automated: no person at Obiter Review reads a submitted report or its review.
- Reports and reviews are permanently deleted once the review is delivered, and nothing is used for training.
Transparency still matters
Using a review tool does not remove the need for openness, and we would not suggest otherwise. The expert in Cornwall Council v M explained her AI use only once questions were raised, and the court then had to unpick how her reports had been produced.
The Civil Justice Council’s October 2026 report recommends a declaration where AI has generated or influenced the substantive content of a report. Speaking at the Bond Solon Expert Witness Conference, Mrs Justice Joanna Smith urged experts not to wait for a formal rule, and to be open about AI use that “may directly or even indirectly generate or influence the content of your evidence”.
A review that prompts an expert to reconsider a passage may, at the margins, influence how it is finally expressed. Not every use of this kind strictly needs reporting under the Academy of Experts guidance, but our advice is simple: say so. A short statement in the report removes any room for doubt, for example:
Before finalising this report I used Obiter Review, an independent quality-assurance service, to check it for internal consistency, procedural compliance and drafting errors. Obiter Review did not draft, amend or suggest wording for any part of this report. The opinions expressed are my own.
Experts should also keep their copy of the review, so that they can explain exactly how it was used if asked. Any expert who uses AI must be able to do that, whatever the tool.
The bottom line
The courts are not hostile to AI. They are hostile to experts handing their judgement to it. Cornwall Council v M shows what happens when AI writes the report; Obiter Review exists for the opposite purpose – to help experts make sure the report they wrote themselves is clear, consistent and ready to withstand scrutiny.
Obiter Review is led by practising consultant forensic psychiatrists who write and give expert evidence themselves, and we use it in our own practice. If you would like to discuss a review, please email [email protected] with a brief outline of the report and the proceedings it is for.
Dr Nicholas Taylor, Co-founder, Obiter Review
Sources
- Cornwall Council v M & Ors [2026] EWFC 306 – full judgment (BAILII) – paragraph numbers in square brackets refer to this judgment
- Bond Solon: Court warns experts against over-reliance on AI in report writing
- Mrs Justice Joanna Smith: Artificial Intelligence and Expert Evidence (October 2026)
