The case that does not exist
On Monday, Legal Futures reported that the Crown Prosecution Service had, for the second time this year, put a fictitious authority before the High Court.
The case was R v J, said to be a 2024 decision, cited three times in the written record of a decision to continue a prosecution. Lord Justice Holgate and Mr Justice Wall called it unsatisfactory but found it did not affect the outcome. In July, the CPS had already admitted putting AI-hallucinated cases before the High Court.
That this happened at the CPS, with its resources and review layers, should tell every small firm something. This is not a problem of naive juniors or careless sole practitioners. It is a problem of unverified text entering documents that carry professional responsibility.
AI found me a precedent!
The question is not "did you use AI"
The Civil Justice Council's final report last week, chaired by Sir Geoffrey Vos, gives a clearer steer than most of the noise. Its starting point: regulation of AI use in preparing court documents remains with legal representatives and their regulators.
The recommendations that matter in practice:
● Witness statements from witnesses of fact must declare that AI tools did not generate or influence their substantive content, so the evidence remains in the witness's own words.
● Expert witnesses should declare AI use in enough detail for the court and the opposing party to scrutinise and test it.
● Guidance for litigants in person needs urgent work.
● Regulators should consider joint guidance, with responsible AI use built into professional training and ethics.
Note the emphasis: on the result of AI use, not whether a tool was opened. The SRA takes the same line. It grants no tolerance for novelty; it asks whether the decisions taken at the time were reasonable.
What this looks like in a small firm
Three habits that need no new software.
Verify every citation against a source you can see. Not a summary, not a search result, not a confident line in a draft. The judgment itself, on a database that holds it. An authority you cannot open is an authority you cannot rely on.
Ask who is responsible, by name. If your name is on it, you own its accuracy. Say so out loud to whoever prepares the first draft, whether trainee, paralegal or system.
Record the check. A one-line note that a citation was verified, and by whom, turns an unprovable assurance into evidence. When the question is did the firm know what it was doing, and can it prove it, the note is the answer.
The one-page policy still does the job
A small firm does not need a technology strategy. It needs six documented decisions and a named person who owns them: which tools are permitted, what data may go into them, who may use them, what must be checked before anything leaves the building, how errors are reported, and who is accountable.
Add one line to your file review checklist asking whether AI was used and how it was verified, and your quality process catches this without a new system.
What this does not mean
I am not arguing that firms should ban AI. Used well it drafts, summarises and organises faster than any previous tool, and refusing to engage is its own risk. The CJC's approach is deliberately risk-based and proportionate.
Nor is this an attack on the CPS. The court said the error did not matter to the outcome. Institutions that admit mistakes publicly behave better than those that do not.
And no policy eliminates error. The aim is that errors are caught by a check rather than by a judge.
The uncomfortable part
The difficult feature of a fabricated authority is how reasonable it looks. Correct formatting. Plausible citation. A holding that fits the argument you were already making. It confirms what you hoped, which is why it slips through.
The defence is not vigilance in the abstract. It is a boring, specific habit: open the source, or do not cite it at all.
Everything else is commentary.
Start with a conversation
If you would like an independent review of how your firm handles verification and AI, including a one-page policy shaped around your practice, I offer a complimentary health check with no obligation. The contact page is an enquiry form. Put "complimentary health check" in the subject and I will pick it up straight away.
Ashley Barwick is a Fellow of the Chartered Institute of Legal Executives and a Law Society accredited Lexcel Consultant. This post is general information and does not constitute legal advice. Fraxinus Business Associates is not authorised to provide legal advice.
Sources
● Legal Futures, "CPS puts another fictitious case before High Court", Neil Rose, 2 November 2026. https://www.legalfutures.co.uk/latest-news/cps-puts-another-fictitious-case-before-high-court
● Legal Futures, "Witnesses of fact and experts 'should have to declare AI use' to court", Neil Rose, 30 October 2026. https://www.legalfutures.co.uk/latest-news/witnesses-of-fact-and-experts-should-have-to-declare-ai-use-to-court
● Civil Justice Council, "Use of AI for the Preparation of Court Documents: Final Report", October 2026. https://www.judiciary.uk
Comments