Skip to content
SafeLegalAI

72 on the record · +1 this week

guidance diffAI in CourtsAI Governance

UK judicial AI guidance: the version history, December 2023 to October 2025

How the Courts and Tribunals Judiciary's AI guidance for judges in England and Wales changed across three versions: Dec 2023, April 2025, and 31 October 2025.

Daman Kaur

The Courts and Tribunals Judiciary of England and Wales was among the first in the world to give its judges written guidance on generative AI. Since December 2023 that guidance has been revised twice. This is the version record: what each edition said, and what changed.

The guidance governs judicial office holders — judges, tribunal members, and their clerks, legal advisers, and support staff. It is not a court rule binding on the parties; it is internal guidance on how the bench itself may use AI.

The three versions

Version Date Status
v1 12 December 2023 Superseded
v2 15 April 2025 Superseded
v3 31 October 2025 Current

Version 1 — 12 December 2023

The first edition set the baseline that has held ever since. Its core requirements:

  • Judicial office holders should have a basic understanding of AI’s capabilities and limitations before using it.
  • Public AI chatbots are a poor way to find new information that cannot be independently verified; they do not provide answers from authoritative databases but generate text predicted to be most likely.
  • All AI output should be checked before use, because it may be inaccurate, incomplete, or misleading.
  • No private or confidential information should be entered into a public AI tool; inputs may become publicly available.
  • A judge is personally responsible for material produced in their name.

Version 2 — April 2025

The April 2025 edition updated and replaced v1. The current published guidance does not reproduce the April 2025 text; what is on the record is that the 31 October 2025 version updates and replaces it, roughly six months later.

Version 3 — 31 October 2025

The current edition, published 31 October 2025, “updates and replaces” the April 2025 guidance. Per the judiciary’s own account, the refresh:

  • expands the glossary of common terms;
  • expands the treatment of bias in training data and of AI hallucinations that generate incorrect or misleading information;
  • adds further advice on confidentiality, including where to report an accidental disclosure as a data incident.

The through-line across all three versions is unchanged: verify everything, keep confidential material out of public tools, and own the output. What has grown is the detail — the risks the guidance names, and the terms it defines.

Why the version record matters

Guidance that changes is guidance that is being enforced against a moving target. The 2023 baseline was written before most of the incidents in our tracker had happened; the October 2025 refresh was written after courts in England, and across the common-law world, had spent two years sanctioning lawyers for exactly the failure the guidance warns about. Reading the versions in sequence shows the judiciary’s risk assessment hardening in real time.

Sources