The judiciary's AI guidance is the clearest public statement of how the bench approaches these tools, and most firms have never read it. First published in December 2023 and refreshed on 31 October 2025, it applies to judicial office holders in England and Wales, together with their clerks, judicial assistants, legal advisers and support staff. It is written for judges, which is exactly why it repays a law firm's attention, because the people who read your filings have been told what to watch for.
What the guidance says to judges
The refreshed edition strengthens three themes. Confidentiality comes first: office holders are reminded not to enter private information into public AI tools, and an accidental disclosure is reported as a data incident like any other loss of court information. Accuracy comes second, with expanded warnings on hallucination, the confident invention of incorrect or misleading material, and on bias carried in from training data. The third theme anchors the other two. Lord Justice Birss puts it plainly: the use of AI by the judiciary must be consistent with the overarching obligation to protect the integrity of the administration of justice. Office holders carry personal responsibility for everything produced in their name, whatever tool helped to produce it.
Since its first edition the guidance has drawn a line between tasks. Summarising large bodies of text and administrative drafting sit on the useful side. Legal research and analysis sit on the other, because a general tool cannot be relied on for authoritative sources, and the guidance has been blunt about fabricated citations from the start.
Judges have been told to look for AI in your work
The part that should sharpen a firm's attention is the direction that judges stay alert to the use of AI by those appearing before them, represented parties and litigants in person alike. The first edition listed tell-tale signs, authorities that cannot be found, unfamiliar citations, American spellings in an English filing. A judge reading your skeleton argument has been briefed on what machine-assisted sloppiness looks like. The hallucination cases that followed, with wasted costs orders and regulatory referrals, show what happens when that primed eye finds something.
The standard your firm should match
Read as a whole, the guidance is a working model of proportionate AI governance from the institution that answers for the administration of justice. It does not ban the tools. It names the risks, keeps confidential information out of public systems, insists on checking, and fixes responsibility on the person whose name the work carries. Substitute solicitor for judge and you have a serviceable first draft of a firm's own position, which is no accident, since the SRA's expectations run along the same lines, set out in our guide to the SRA and AI.
The current edition is short, readable and free on the judiciary’s AI guidance page.
If the bench holds itself to confidentiality, checking and named responsibility, your firm can expect to be measured against nothing less. We turn that standard into a working policy and daily habit inside small firms: start with a conversation.