Why chambers need AI rules before barristers use tools on client work
Before AI is used for client work, chambers should agree which tools and data are permitted, how legal authorities and facts will be checked, and who is responsible for the final work. Record material uses without keeping more client content than the matter requires.
The policy does not need to be long. It does need to be specific enough that a barrister, clerk, pupil supervisor or chambers manager can tell which tools are approved, which data is prohibited, what records must be kept and who checks the work before it is used.
The safest approach is a short policy, mandatory records and independent verification
Agree a policy, a proportionate material-use record and source-checking rules for advice, pleadings, research notes and correspondence. These are proposed chambers controls; identify applicable professional obligations separately rather than treating this article as their source.
Confirm the exact BSB guidance, its date and applicability before attributing requirements to it. For the proposed local record, identify the tool, data category, material output, source checks and final professional judgement. Treat this as a proposed local record, not a BSB requirement to log every workflow.
Review AI risks in your professional services workflow
What the chambers AI policy should cover
A useful policy should be written around real work, not abstract technology. Start with approved and prohibited uses.
- Approved tools: name the AI tools or embedded features that may be used, including any approved settings for retention, training, privacy and access control.
- Prohibited data: state when client names, privileged material, special category data, witness information, unpublished advice or case strategy must not be entered.
- Permitted use cases: separate low-risk tasks such as first-pass summarisation of non-confidential public material from higher-risk work such as legal research, drafting, evidence analysis or client communications.
- Human responsibility: make clear that AI output is never the final professional judgement. The barrister remains responsible for accuracy, independence, confidentiality and duty to the court.
- Escalation: set a route for questions about unusual tools, client objections, suspected data exposure, hallucinated authorities or conflicts with court directions.
What the audit log should record
The audit log should be light enough to use and detailed enough to reconstruct the decision. A spreadsheet, matter management field or secure form can work if it is consistently used.
- matter or work reference, without unnecessary client-identifying detail in the log itself;
- date, user and tool used;
- purpose of use, such as research support, drafting support, chronology review or administrative summary;
- data category entered, such as public law, anonymised facts, confidential client facts or personal data;
- whether any prompt or output history is retained, and where;
- verification steps completed;
- final reviewer or responsible barrister;
- incidents, anomalies, hallucinated cases or corrections.
Before sensitive client work enters a tool, assess confidentiality and data protection and record the decision. Agree what evidence must be retained, who may access it, where it is held and its retention period against the matter's requirements. A data category or controlled source reference may be preferable to copying a complete prompt or client file; neither is automatically sufficient. No BSB prompt-history or audit-log requirement is established here.
Verification rules for barristers using AI
The verification rule should be simple: no AI output is relied on until it has been checked against independent sources. This matters most where AI touches law, facts, authorities, quotations, chronology, calculations, evidence or client advice.
- Legal propositions: check against primary sources, current legislation, authoritative guidance or trusted legal databases.
- Cases and citations: verify that every authority exists, says what the output claims and remains good law.
- Facts: trace claims back to client documents, witness evidence, disclosed material or agreed instructions.
- Bias and protected characteristics: review whether prompts or outputs could introduce discriminatory assumptions.
- Client-ready drafting: review tone, privilege, confidentiality, accuracy and strategic judgement before anything leaves chambers.
A good test is whether the barrister could explain the check to a supervisor, client, regulator or court. If the answer is no, the output is not ready.
How chambers should manage governance and training
Nominate an owner for the approved-use list, vendor-setting reviews and staff guidance. Use practical training sessions to test the rules, and confirm any applicable CPD requirements separately; this article does not establish CPD credit or a mandatory AI-training programme.
Training should cover confidentiality, prompt hygiene, verification, bias, record keeping and when to avoid AI altogether. It should also cover the ordinary reality of AI tools: they can sound confident while being wrong, incomplete or unsuitable for a legal context.
Set up practical AI governance for your firm or chambers
A practical starting checklist
- List the AI tools already being used in chambers, including embedded features in research, document, email and practice management systems.
- Classify use cases by risk: admin, public research, anonymised drafting support, confidential client work and court-facing work.
- Decide which tools are approved, restricted or banned.
- Set default privacy, retention and training settings for each approved tool.
- Create a one-page AI use log for material client work.
- Write verification rules for law, facts, citations, calculations and client communications.
- Define an incident route for accidental disclosure, wrong authorities or unsafe outputs.
- Agree the policy review schedule against the uses and applicable requirements. Quarterly review is an example, not an established minimum; reassess relevant vendor, regulatory or court-guidance changes when they occur.
Conclusion
Test the proposed rules against a sample matter: identify permitted data, trace the source checks and record the responsible barrister's review. Assign any gap before relying on the output. A policy or log alone does not establish compliance or protection for clients or the court.
Build an AI policy and audit trail that your team can actually use
Need a safer AI route for your firm?
If this article reflects a live decision in your practice, the useful next step is to map the workflows, confidentiality risks, supplier controls and governance gaps before tools spread informally.
