The Bar Council on ChatGPT and LLMs: what barristers should know
Considerations when using ChatGPT and generative AI software based on large language models
- Issued by
- The Bar Council of England and Wales
- Published
- 30 January 2024
- Updated
- 25 November 2025
- We reviewed
- 17 July 2026
The Bar Council’s considerations document on ChatGPT and LLM-based tools (January 2024, reviewed November 2025) does not prohibit generative AI — it sets out how to use it without breaching core duties. The sharpest lines are about data: be extremely vigilant about sharing legally privileged or confidential information with an LLM system, and do not input personal data into prompts. The November 2025 review added the Ayinde fabricated-citations judgments, connecting careless AI use directly to court sanctions. Note: the document describes itself as assistance, not formal BSB guidance.
What the guidance says
“Be extremely vigilant about sharing with a generative LLM system any legally privileged or confidential information”
“anything that a user types into the system may be used to train the software and might find itself repeated verbatim in future results”
Personal data should not be entered into prompts at all; the document suggests synthetic data as a way to avoid UK GDPR breach.
The November 2025 review added the Ayinde v Haringey judgments on fabricated AI citations — the courts are now part of this story, not a hypothetical.
What this means for your firm
- For chambers and barristers, the input side (what reaches the model) is treated as seriously as the output side (hallucinated citations).
- "Vigilance" as a human-only control fails under deadline pressure — a technical mask-before-send step makes the vigilant path the default path.
- The synthetic-data suggestion maps directly to masking: structurally faithful placeholders preserve the legal question while removing the identifying content.
- This is an assistance document, not binding BSB guidance — but it is what a disciplinary tribunal would expect a barrister to have read.
Guidance → control, line by line
How each expectation maps to a NeutralAI control. The full cross-regulator table lives on the UK guidance map.
Extreme vigilance with privileged/confidential information
Identifying and confidential specifics are removed from the prompt automatically — vigilance becomes a system property instead of a memory test.
Never input personal data into prompts
Names, addresses, NI and NHS numbers, court references, and other UK identifiers are detected and replaced before the text leaves the browser.
Prefer synthetic stand-ins over real data
Placeholders like <PERSON_7K9X> act as consistent synthetic stand-ins; the real values are restored only in your environment after the response returns.
Common questions
Is the Bar Council document binding on barristers?
No — it explicitly states it is not formal guidance for BSB Handbook purposes. It is an assistance document from the Bar Council’s IT Panel. In practice, it sets the expectations a regulator or tribunal would assume a careful barrister knows.
Can barristers use ChatGPT for drafting at all?
Yes, the document contemplates legitimate use. The constraints are on what goes in (no privileged, confidential, or personal data) and on verifying what comes out (the Ayinde judgments on fabricated citations are now cited in the November 2025 review).
How does masking relate to the "synthetic data" suggestion?
Masking automates it: real identifiers are replaced with structurally faithful placeholders before the prompt leaves your machine, which is exactly the synthetic stand-in pattern the document suggests — without asking the barrister to build synthetic examples by hand.
This page summarises third-party guidance for convenience and is not legal advice. Summaries can go stale — always read the original at the source link above before relying on it. Last reviewed: 17 July 2026.
See what this control looks like in practice
The AI Confidentiality Checklist walks through usage discovery, exposure, policy, controls, and evidence in about 20 minutes — or bring one low-risk workflow to a live review.
The control
detect → mask → send → restore → audit
Reversible vault, 15-minute TTL. The model only ever sees placeholders.