On August 25, 2026, the Bar Council published a blog by Anna Thomas MBE arguing that the way artificial intelligence changes legal work depends on choices the profession makes now. That framing is right—and it demands more than warnings. Legal AI adoption needs proof of value and control in day‑to‑day practice, not just policy statements.
What the Bar Council is saying about legal AI adoption
Thomas, co‑founder of the Institute for the Future of Work, writes that use of AI inside chambers and firms is growing but poorly understood. Citing a LexisNexis survey, the Bar Council blog reports that 43% of barristers already use AI for legal work. Former Chair Barbara Mills KC has called the growth of AI tools “inevitable” and “fast‑paced.” The piece also flags recent case law: in Ayinde v London Borough of Haringey, fabricated authorities reached the court record, prompting a Divisional Court warning about AI misuse and its effect on the administration of justice (Bar Council).
The headline message is accountability. The Bar Council’s published guidance stresses that barristers remain responsible for their work product; a hallucinated citation is not a software error, it is professional misconduct risk. That aligns with long‑standing obligations in the Bar Standards Board Handbook and with judicial expectations on accurate authorities, reflected in the judiciary’s Practice Direction on citation of authorities. The warning is clear. The next step is turning it into operational habits.
Where the real risk sits: authority, not algorithms
The blog’s most important implication is easy to miss: the system’s weak point is legal authority, not the model. A model can draft at speed, but only counsel can stand over what goes before a court. The Ayinde episode shows how fast an unchecked paragraph can become a filing that corrodes trust. If legal AI adoption pushes more unverified text into workflows, risk compounds at the busiest moments—late edits, urgent mentions, skeleton arguments under time pressure.
The fix starts earlier than citation checks at the end. It requires matter‑level rules about when AI may be used, how output is tagged for review, and who signs off. It also means structuring reviews around the authority chain: primary law, secondary sources, and commentary. An assistant that suggests a case isn’t the problem; a process that treats the suggestion as a source is.
A practical, auditable plan chambers can run this term
The Bar Council’s call for agency over change invites specifics. Here is a simple plan a set of chambers can adopt within a fortnight to make AI in legal work safer and more useful—and to produce evidence that it is.
- Create an AI use register. For each matter, record whether AI assisted drafting, research, translation, or none. Keep the register at practice‑group level and name a senior reviewer.
- Mark AI‑touched text. Require a short note at the top of any draft that includes AI‑generated content. The note should name the task (e.g., “issue spotting”), list any authorities AI suggested, and assign a human verifier.
- Adopt a two‑pass authority check. First, verify every citation against primary sources or trusted databases. Second, assess relevance against the live issues and the hearing’s jurisdiction. No exceptions for “well‑known” cases.
- Log four metrics on every AI‑assisted task: verification time, citation error rate, instances where AI surfaced a useful but unknown authority, and any confidentiality concerns raised by the reviewer.
- Set red lines. For now, bar AI use on without‑notice applications, criminal sentencing submissions, and any filing that relies on unpublished or sensitive data, unless a silk or supervisor signs off.
This plan answers two questions the Bar Council blog leaves open: where to focus controls, and how to prove they work. It converts broad accountability into checks a clerk can schedule and a leader can review. It also gives insurers and regulators something concrete: not a pledge, but a record.
What to measure to steer legal AI adoption
Metrics shape behaviour. To steer legal AI adoption toward value and away from risk, track measures that link directly to duties the courts and regulators already expect.
- Authority accuracy: percentage of AI‑suggested citations that survive verification. Target 100% on filings, 95%+ on internal memos. Report monthly by team.
- Human time saved where accuracy holds: minutes saved per page after a clean authority check. If speed rises but accuracy drops, the tool isn’t ready.
- Complaint and query rate: number of client or court queries about sources tied to AI‑assisted work. Zero is the only acceptable figure on filed material.
- Confidentiality flags: count of matters where an AI tool’s terms or data flows were judged unsuitable. Use this to refine procurement and briefing.
Publish the topline results to members each term. If a tool’s authority accuracy is low, limit it to ideation or plain‑English summaries. If time saved and accuracy both rise, expand the use case. In other words, let numbers—not hype—decide the next step.
Policy, insurance, and the cost of getting this wrong
No one wants a repeat of Ayinde. But the real cost of a single failure often comes later: higher excesses at renewal, tighter supervision orders, and reputational drag that outlasts a news cycle. According to the Bar Council blog, the profession must keep agency over how technology changes work. If chambers can show measured controls and low error rates, they keep that agency with evidence that underwriters, judges, and clients will accept.
There’s another benefit: better training. Data from your AI use register will show where juniors spend verification time and which practice areas win most from these tools. That lets Heads of Group tune training sessions to the real pinch points and cut busywork without blurring responsibility. It also gives clerks a sober way to price matters that involve AI‑assisted drafting, because you’ll know the verification time it typically adds or saves.
The Bar already understands process and proof. Applying both here isn’t a culture shift; it’s craft. Start small, publish the metrics, and adjust. If legal AI adoption is rising—and the Bar Council says it is—the profession can still decide what rises with it: trust, or risk. For more on this, see bloomberg.com and nytimes.com.
Related reading: AI in Education • Data Privacy • AI in Society