The BSB’s new AI guidance: what actually changed for barristers
On 18 May 2026 the Bar Standards Board’s Guidance on the use of Artificial Intelligence and Other Technologies came into force. It rewards a full read, but here is what it actually does — and the one shift practitioners shouldn’t miss.
1. It isn’t a new regime — it maps AI onto duties we already owe. The BSB hasn’t invented a separate AI rulebook. It has read AI into the existing Core Duties: competence (CD7), the client’s best interests (CD2), honesty and integrity (CD3), confidentiality (CD6), the duty to the court (CD1) and practice management (CD10). The practical effect is that there is no “AI exception” to anything we already do.
2. Competence is expected even if you never touch the tools. This is the line many will skim past. You are expected to understand AI’s impact whether or not you use it yourself — because opposing counsel, clients and the courts will, and because the admissibility of AI-generated evidence and prompt histories is becoming a live issue.
3. Using AI is outsourcing — so it’s your name on it. The guidance treats AI use as outsourcing under rC86. Translation: any inaccuracy the tool introduces remains yours. That echoes Ayinde and Al-Haroun [2025] EWHC 1383 (Admin), where the Divisional Court used its Hamid jurisdiction after fabricated authorities were put before the court. “The model drafted it” is not mitigation.
4. Free, public tools are largely off-limits for client work. Because their terms typically let providers retain and train on user inputs, the BSB’s view is that they are unlikely to meet confidentiality obligations — and uploading client material to an open tool can waive privilege. Use them, if at all, only for the lowest-risk, non-client tasks.
5. A risk-based approach, with a usable matrix. The guidance grades risk across three axes — application, use and technology — from low (admin, grammar) through medium (legal research in secure tools) to high (court submissions, vulnerable clients, drafting). Agentic features, which let a system act on its own output, are flagged as high risk and to be approached “with absolute caution.”
6. Transparency is contextual, not automatic. You must tell clients where AI materially affects the nature or scope of your service (rC19), and you must always answer the BSB or the court honestly if asked. Routine, de minimis use generally needn’t be flagged.
The bigger picture
What’s quietly significant is the posture. This is neither a prohibition nor cheerleading — it’s a competence-and-accountability framework. The Bar is saying that judgment, not abstinence, is the professional standard. The practitioners who come unstuck won’t be the ones who used AI; they’ll be the ones who couldn’t explain how, couldn’t show they’d verified the output, or assumed the rules paused because a model was in the loop.
For those of us in cross-border and offshore work there’s an added wrinkle: data moving between jurisdictions, and clients and overseas counsel who may already be running matters through tools we would never put client data into. Knowing what to ask is now part of the brief.
The guidance plainly favours those who already think in terms of governance — a clear policy, audit logs, and a simple “what we will and won’t use AI for” playbook. That isn’t bureaucracy. It’s the evidence that you took a reasonable, informed approach, if anyone ever asks.
Worth your time. The full guidance sits on the BSB Handbook site.
